Safeguard
Supply Chain Security•

How Drones Get Hacked and How to Secure Them

Drone security explained: how drones get hacked through apps, MAVLink links, firmware, GNSS spoofing and Remote ID, and how to secure them.

Safeguard Security Team
63 min read

Drones get hacked in five main ways. Attackers take over the accounts and cloud services that store flight logs and imagery. They send commands over radio links that do not check who is talking, which is the default for the open MAVLink protocol. They exploit bugs in drone firmware and the parsers that read radio, sensor and app input. They fool or block satellite navigation with GNSS spoofing and jamming. And they read the identity and location data that drones broadcast. Almost every documented case was fixed by an update from the company that wrote the software, and most of the long-term risk sits in the drone software supply chain.

That is the short answer. This guide is the long one. It covers consumer and commercial drones only: camera drones, survey and inspection aircraft, agricultural sprayers and the hobby and maker builds that run open source autopilots. It explains what software runs in a drone system and where an attacker can reach it, then walks through the documented research: the leaked DJI key from 2017, Check Point's 2018 DJI account takeover, the MAVLink authentication CVEs, the Synacktiv analysis of the DJI GO 4 Android app, the 2023 Ruhr University Bochum and CISPA work on DJI's DroneID and firmware, and PX4's security advisories. It covers GNSS spoofing and jamming, data privacy, companion apps, Remote ID, enterprise fleets and the drone software supply chain, and ends with hardening checklists for hobby pilots, enterprise fleet operators and makers.

It is written from the attacker's point of view for defenders. For each attack you will find how it works and why, what the attacker needs, real examples, the impact, how to detect it and how to defend against it. You will not find exploit code, tool command lines or step-by-step instructions. Where a tool or project is named, it is named as context for authorised testing.

A note on the law. Testing drones, controllers, apps or cloud services that you do not own, or have no written permission to test, is illegal in most countries. Interfering with radio or satellite navigation signals is also illegal in most places, even on your own equipment, outside a licensed test range or shielded lab. Everything in this guide is for protecting your own aircraft, your organisation's fleet with authorisation, or the products you build. The researchers cited here worked on their own hardware and disclosed what they found to the vendor.

This post is part of a device security series. The companion guide on how Wi-Fi gets hacked and how to secure it covers the radio that many small drones use for control and video. The guides on how phones get hacked and how laptops get hacked cover the devices that run drone apps and ground control software. Our posts on fabless chip designers and the software supply chain and how a laptop is built explain where the chips and vendor SDKs inside flight controllers and radios come from.

Contents

  1. How does a drone work, and what software does it run?
  2. What is the attack surface of a drone?
  3. What does an attacker need to hack a drone?
  4. Which drone security research should you know about?
  5. How are drone command links attacked?
  6. What did researchers find in DJI drones and services?
  7. How do PX4 and ArduPilot handle security?
  8. How do GNSS spoofing and jamming affect drones?
  9. What data does a drone expose?
  10. What are the risks in drone companion apps?
  11. What does Remote ID mean for privacy?
  12. How should you weigh these risks?
  13. How do you detect attacks on drones?
  14. Why is drone security a supply chain problem?
  15. How should hobby pilots secure their drones?
  16. How should businesses secure a drone fleet?
  17. How should drone makers build secure products?
  18. How Safeguard helps teams that build drone software
  19. The takeaway
  20. Frequently asked questions
  21. Sources

How does a drone work, and what software does it run?

A drone system is four layers of software. The aircraft runs flight controller firmware on a small real-time operating system. Radio links carry stick commands, telemetry and video. A ground station, usually a phone app or desktop program, plans and monitors flights. And a cloud side stores accounts, flight logs, media and updates.

People tend to think of a drone as the aircraft. From a security point of view it is a distributed system, and the aircraft is only one node in it.

Layered diagram of a drone system showing the cloud (accounts, log and media sync, fleet management, updates and geofence data), the ground (remote controller, ground control app, mission planning, local storage), the links (RC control link, telemetry link, video downlink, Wi-Fi) and the aircraft (flight controller, sensors, payload, companion computer and Remote ID)

Figure 1. The four software layers of a consumer or commercial drone system.

The aircraft

At the centre is the flight controller, a small computer that reads the sensors and drives the motors hundreds of times a second to keep the aircraft stable. It runs firmware, the software stored on the device itself, usually on top of a real-time operating system (RTOS), an operating system built to respond to events within strict time limits. The open source PX4 autopilot runs on NuttX, according to its Wikipedia entry, and ArduPilot's hardware abstraction layer includes a ChibiOS port for STM32-based boards, an ESP32 port and a Linux port, as its developer introduction explains. Large consumer drone makers such as DJI write their own firmware, which spans several processors: the flight controller, the camera and video system, and the radio transceiver.

The flight controller depends on sensors. An inertial measurement unit (IMU) measures rotation and acceleration. A compass gives heading. A barometer estimates altitude. And a GNSS receiver (global navigation satellite system, the umbrella term for GPS, Galileo, GLONASS and BeiDou) gives position. Many aircraft add downward cameras for optical flow, rangefinders and obstacle sensors.

Then there is the payload: a camera on a gimbal that writes photos and video to an SD card, a multispectral sensor on a survey drone, a spray system on an agricultural drone. Bigger commercial aircraft often carry a companion computer, a Linux board that runs higher-level software such as mapping, object detection or ROS 2 (the Robot Operating System), and talks to the flight controller over a serial or network link. Many drones also carry a Remote ID broadcaster, which announces the drone's identity and position over Bluetooth or Wi-Fi.

The links

A drone usually has several radio links, sometimes multiplexed onto one radio:

  • The RC control link carries stick commands from the remote controller. Hobby systems use serial RC protocols such as CRSF.
  • The telemetry link carries status and commands between the aircraft and the ground station. In the open ecosystem that protocol is MAVLink, a lightweight message protocol used by PX4, ArduPilot and most open ground stations. DJI uses its own protocols and its proprietary OcuSync radio system; the NDSS 2023 paper notes that, according to DJI, OcuSync's uplink and downlink are AES-256 encrypted.
  • The video downlink carries the live camera feed.
  • Many small and toy drones simply create a Wi-Fi network that a phone joins, and run control, telemetry and video over it.

The ground

The remote controller is a computer too, often running its own embedded OS or Android. The ground control station (GCS) is the software the pilot uses: a phone or tablet app for consumer drones, or programs such as QGroundControl and Mission Planner for open autopilots. QGroundControl ships as a Windows installer, a Linux AppImage and builds for other platforms, per its install guide, and Mission Planner ships as a Windows .msi installer, per ArduPilot's documentation. The GCS stores flight logs, cached maps and often media.

The cloud

The cloud side handles sign in, flight log and media sync, fleet management with live views, firmware and app updates, and geofence databases that define no-fly zones. For a consumer pilot this is often invisible. For an enterprise fleet it is the system of record.

Every one of those layers runs software someone wrote, much of it open source, and every layer can carry a vulnerability.

What is the attack surface of a drone?

A drone's attack surface spans five zones: the air (command links, Wi-Fi, GNSS and broadcast IDs), the aircraft (firmware parsers, debug ports, storage and companion computers), the ground (apps and ground station software), the cloud (accounts and synced data) and the supply chain (open source autopilots, RTOS, chip SDKs and build pipelines).

The attack surface is every place where an attacker can feed input into a system or pull data out of it. For a drone it looks like Figure 2.

Map of five attack surface zones: over the air (unauthenticated command links, open Wi-Fi, GNSS spoofing and jamming, broadcast IDs, readable telemetry), on the aircraft (parsers, debug ports and bootloaders, SD card, companion network, geofence and Remote ID tamper), on the ground (app permissions and SDKs, self update paths, ground station hosts, stored data, shared tablets), in the cloud (account takeover, synced logs and media, fleet portals, leaked keys, third party log services) and in the supply chain (PX4 and ArduPilot, RTOS and chip SDKs, generated MAVLink libraries, CI workflows, vendor forks)

Figure 2. Where a drone system can be attacked. Each zone has documented cases in this guide.

A few points stand out.

First, the radio links are the defining feature. A drone is a computer that you deliberately put out of reach and talk to over the air. Anything that accepts commands over that air gap is reachable by anyone within radio range who speaks the protocol.

Second, the phone matters as much as the aircraft. On consumer drones the phone app is the pilot's interface, and the NDSS 2023 researchers showed that almost all of the commands they found in DJI firmware could be sent over the air from the smartphone connected to the controller. A compromised phone is a compromised drone.

Third, the cloud holds the most sensitive data. Flight logs record where you flew, where you took off and where every photo was taken. If they sync to an account, an account takeover is a location history breach.

Fourth, the supply chain is wide. A single drone can contain an open source autopilot, an RTOS, vendor radio firmware, a Linux companion image, a mobile app with third-party SDKs and a cloud backend. Each comes from a different place, and each has its own vulnerability history.

What does an attacker need to hack a drone?

It depends on the layer. Cloud and app attacks need only an internet connection and a weak account or app. Link attacks need radio range and the right protocol knowledge. Firmware attacks usually need research time and a drone to study. GNSS attacks need transmitters that are illegal to operate in most places without a licence.

Attack classWhat the attacker needsTypical reach
Account and cloud takeoverA flaw in the vendor's web services, phished or reused credentials, or a leaked keyGlobal
Open Wi-Fi drone takeoverA laptop or small computer within Wi-Fi range of an unprotected droneTens to hundreds of metres
Unauthenticated MAVLink commandsRadio range and a compatible radio on the same link, or access to the network the link runs onRadio range, or anywhere on that network
Firmware exploitationA copy of the firmware, a drone to test on, reverse engineering time, then a delivery path (radio, app or update)Varies, often radio range or a compromised phone
Malicious or compromised appA path onto the pilot's phone or laptopGlobal
GNSS jammingA transmitter on GNSS frequencies (illegal to operate in most countries)Line of sight
GNSS spoofingA transmitter that can produce convincing counterfeit signals and track the targetLine of sight, harder to do well
Reading broadcast IDsA phone app for Remote ID, or a software-defined radio for proprietary broadcastsRadio range
Supply chain compromiseAccess to an upstream project, a CI pipeline, a signing key or a vendor update channelEvery drone that ships the code

Two observations. Most of the cheap, wide-reach attacks are not about flying at all; they are about data. And the attacks that do affect flight safety mostly need radio proximity, which limits who can attempt them and where.

Which drone security research should you know about?

The landmark public cases are SkyJack (2013), the DJI SSL key found on GitHub (2017), Check Point's DJI account takeover (2018), the MAVLink authentication CVEs and the DJI GO 4 analysis (2020), the NDSS 2023 DroneID and DJI firmware research, and the PX4 advisory batches of 2023 and 2026. Nearly all were disclosed to the vendor and fixed.

Timeline from 2013 to 2026: SkyJack and GPS spoofing of a yacht (2013); DJI SSL private key found on GitHub (2017); Check Point DJI account takeover (2018); MAVLink CVE-2020-10282 and 10283 and the DJI GO 4 Android analysis (2020); NDSS DroneID decoding and 16 DJI firmware bugs plus PX4 parser CVEs (2023); Galileo OSNMA operational (2025); PX4 advisories including CVE-2026-32709 (2026)

Figure 3. A timeline of documented drone security research, from 2013 to 2026.

The pattern is consistent. Early work showed drones trusting anyone on the radio. Later work moved to the cloud and apps, where the data lives. The most recent work targets firmware and the open source autopilots most non-DJI drones build on.

How are drone command links attacked?

Drone command links are attacked by joining or impersonating the link. Small drones that run on open Wi-Fi can be taken over by anyone in range. Open autopilots speak MAVLink, which in version 1 has no authentication at all and in version 2 authenticates only when message signing is turned on. Proprietary links are harder to reach but not immune.

Open Wi-Fi drones: SkyJack (2013)

In December 2013 security researcher Samy Kamkar published SkyJack, a drone that hunted for other drones. It targeted the Parrot AR.Drone 2, which created an open Wi-Fi network for its controlling phone. Conceptually the attack was simple: find a drone by the manufacturer's prefix in its wireless address, force the real pilot's phone off the network, then join as if it were the pilot and send commands.

Why it worked: the drone's network had no password, and the drone accepted commands from whoever was connected. What the attacker needed: a Wi-Fi radio in range and software that spoke the drone's control protocol. Impact: full control of the target drone and its camera. Detection: the pilot's app loses connection while the drone keeps flying under someone else's control. Defence: never ship or fly a drone on an open network. Makers should use WPA2 or WPA3 with a unique per-device passphrase and pair the controller to the aircraft. Pilots of Wi-Fi drones should set a strong network password where the drone supports one.

Toy and entry-level Wi-Fi drones still exist, and the companion guide on how Wi-Fi gets hacked covers the attacks on Wi-Fi itself.

MAVLink without authentication: CVE-2020-10282

MAVLink is the most widely used open drone protocol. Version 1 has no way to tell a legitimate ground station from anyone else. The CVE Program record for CVE-2020-10282 says it plainly: the protocol "presents no authentication mechanism on its version 1.0 (nor authorization)," which leads to identity spoofing, unauthorised access and person-in-the-middle attacks. A follow-up record, CVE-2020-10283, describes how an attacker could try to push a ground station and autopilot back to version 1 during version negotiation and so sidestep version 2's authentication.

What does an unauthenticated link let an attacker do? PX4 lists it in its MAVLink security hardening guide. On an unsecured link, an attacker within communication range can run shell commands, read, write or delete files, change any flight parameter, upload or overwrite missions, arm or disarm motors, terminate the flight, trigger an emergency landing with a spoofed battery message, and reboot the vehicle. PX4 states: "Until the link is secured, there is no distinction between a legitimate GCS and an unauthorized sender."

Why it works: MAVLink was designed as a lightweight, low-latency protocol for trusted links. Security was left to the integrator. What the attacker needs: a radio on the same frequency and protocol as the telemetry link, or access to the IP network the link runs over (a common setup when a companion computer or a 4G modem bridges MAVLink to the internet). Impact: from changed parameters to a crashed aircraft. Detection: unexpected parameter changes, mode changes or mission uploads in the flight log; a second system ID appearing on the link; telemetry from an unknown ground station. Defence: secure the link below MAVLink (an encrypted radio, a VPN or IPsec), enable MAVLink 2 message signing, or both.

What MAVLink 2 message signing does and does not do

MAVLink 2 added message signing, which the MAVLink guide describes as letting a system "verify that messages originate from a trusted source." A signed packet carries an extra 13 bytes: an 8-bit link ID, a 48-bit timestamp in 10 microsecond units since 1 January 2015, and a 48-bit signature. The signature is the first 48 bits of a SHA-256 hash of a 32-byte secret key combined with the packet and timestamp. Because timestamps must always increase, a recorded packet cannot simply be replayed.

Diagram comparing an unsigned MAVLink 2 frame (header, payload, checksum) with a signed frame that adds a 13-byte signature made of an 8-bit link ID, a 48-bit timestamp and a 48-bit signature, with a note that the signature is the first 48 bits of SHA-256 over a 32-byte secret key and the packet, and that the payload is not encrypted

Figure 4. MAVLink 2 signing adds authentication and replay protection, not encryption.

Signing has real limits, and every operator should understand them:

  • It does not encrypt. The MAVLink FAQ says: "MAVLink does not provide message encryption." ArduPilot's Mission Planner signing page is equally direct: signing "does NOT encrypt the data." Telemetry, including position, stays readable to anyone who can receive the link.
  • It uses one shared key. PX4's message signing page explains that a single secret key is shared by the ground station, the autopilot and any companion computer, camera or gimbal. If any one of them leaks the key, every link is exposed.
  • Some messages are accepted unsigned. The MAVLink specification suggests that RADIO_STATUS packets are always accepted without signing. PX4's scope document lists HEARTBEAT and RADIO_STATUS on every link, plus traffic receiver messages on a link configured for them.
  • The first key can be set by anyone. PX4 warns that until a key is provisioned "any peer that can reach a link can provision one," which would lock out the real ground station. The fix is to provision the key over USB before the vehicle ever goes near an untrusted link.
  • Physical access resets it. On PX4 the key lives on the SD card, and removing the key file disables signing. That is a deliberate recovery path, and it means physical access defeats signing.

Signing is still the single most valuable setting most MAVLink operators are not using. Where a link also needs confidentiality, put it inside an encrypted radio or VPN as well. Proprietary links such as DJI's OcuSync raise the bar because they are undocumented, but obscurity is not a control, as the next section shows.

What did researchers find in DJI drones and services?

Three pieces of research stand out. Check Point showed in 2018 that a web flaw could hand an attacker a DJI account with its synced flight logs, photos and live fleet views. Ruhr University Bochum and CISPA showed in 2023 that DroneID was unencrypted and found 16 firmware vulnerabilities. DJI fixed the reported issues in both cases.

DJI is the market leader (the NDSS 2023 paper puts its share at 94 percent), so it has drawn the most research. Its own Trust Center says it has run a bug bounty since 2017 with rewards from 50 to 30,000 US dollars, and points to independent audits of its products. The cases below are all public and were fixed.

The 2017 SSL private key on GitHub

During the launch of DJI's bug bounty in 2017, researcher Kevin Finisterre and colleagues found that the private key for DJI's SSL certificate had been accidentally published on GitHub, giving access to sensitive customer information on DJI's servers, as The Verge reported. CyberScoop's account describes a server holding customer contact information and user flight records. The disclosure turned into a public dispute over the bounty's terms, which is a story about process. The technical lesson is simpler: a credential committed to a public repository is a breach waiting to be found, and secrets scanning across every repository is the control that catches it.

Check Point's DJI account takeover (2018)

In November 2018 Check Point Research published a vulnerability that would have given an attacker a DJI user's account without the user knowing. The chain started on the DJI Forum: a cross-site scripting flaw (XSS, where a website runs attacker-supplied script in a visitor's browser) let a planted link steal the cookie DJI used to identify the user. That cookie could then be used to obtain tokens for DJI's other services.

The impact is what makes this case useful. According to Check Point, an attacker could have reached:

  • flight logs, photos and videos, if the user had synced them with DJI's cloud servers. Check Point notes that flight logs "indicate the exact location of a drone during its entire flight";
  • a live camera view and map view during flights, for users of DJI's FlightHub fleet management software;
  • the user's profile information.

Check Point notified DJI in March 2018. DJI patched the vulnerability, classified it as high risk but low probability, and said there was no evidence it had been exploited by anyone other than the researchers.

Why it worked: one web flaw on a low-value site (a forum) yielded a credential that the high-value services (cloud sync, fleet management) also trusted. What the attacker needed: a planted link and a logged-in victim who clicked it. Detection: sign-in alerts from new devices, unexpected sessions in account settings, unexplained access in fleet audit logs. Defence: for vendors, isolate authentication between services, mark session cookies HttpOnly and Secure, and fix XSS. For pilots, keep only the data in the cloud that you need there.

DroneID and 16 firmware vulnerabilities (NDSS 2023)

In 2023 a team from Ruhr University Bochum and the CISPA Helmholtz Center for Information Security published Drone Security and the Mysterious Case of DJI's DroneID at the NDSS symposium. The paper has two halves.

DroneID. DJI drones broadcast a tracking protocol called DroneID, designed so that authorised parties can see the position of the drone and its operator. By reverse engineering DJI firmware and the radio signals, the team built a decoder from inexpensive off-the-shelf hardware and showed "that the transmitted data is not encrypted, but accessible to anyone, compromising the drone operator's privacy." They published a proof-of-concept receiver in the RUB-SysSec DroneSecurity repository, tested with a DJI Mini 2 and Mavic Air 2.

Firmware. Using reverse engineering, a new fuzzer built for DJI's DUML command protocol, and hardware analysis, the team found 16 vulnerabilities "ranging from denial of service to arbitrary code execution." Fuzzing means feeding a program large volumes of malformed input to see what breaks. The university's press release explains that some bugs allowed an attacker to change the drone's serial number or log data, override the mechanisms that stop drones flying over airports and other restricted areas, and crash drones in mid-air. The paper adds that 14 of the bugs could be triggered remotely via the operator's smartphone. It also describes a firmware signature bypass on the radio chip: configuration files that were by design neither signed nor encrypted were read by the bootloader and applied after the firmware signature check, which allowed arbitrary changes to firmware code.

The tested models included the DJI Mini 2, Mavic Air 2, Mavic 2 and Mavic 3. The press release states that DJI was informed of all 16 vulnerabilities before publication and that "in the course of the responsible disclosure process the manufacturer has fixed these issues."

Why it matters for defenders: the research shows three general lessons. A signature check protects only what it covers, so anything loaded after it needs its own integrity check. A drone's command protocol is attack surface even if it was meant for internal use, because a phone app can reach it. And a broadcast that is not encrypted is public, whatever the documentation implies.

The wider modding scene

A long-running hobbyist community also modifies DJI firmware, for example through the open source dji-firmware-tools project. For an enterprise, that is a reason to verify that fleet aircraft run the official firmware you expect.

How do PX4 and ArduPilot handle security?

PX4 and ArduPilot are the two main open source autopilots. Both treat the radio link as the integrator's responsibility, both ship MAVLink signing and signed firmware options, and PX4 publishes security advisories through GitHub, including a 2026 batch with a MAVLink FTP path traversal, CVE-2026-32709.

Most drones that are not DJI products run PX4 or ArduPilot, or a fork of one of them. That includes many commercial survey, inspection, agricultural and delivery aircraft. PX4, MAVLink, QGroundControl and MAVSDK sit under the Dronecode Foundation, a Linux Foundation collaborative project.

Where the security boundary is

Both projects now publish clear statements of what they do and do not secure, and anyone deploying them should read both.

ArduPilot's security policy says its security boundary "extends up to, but does not include, the physical communication interface: the radio link, network cable, USB cable and so on. Securing the command and control (C2) link and any network on the vehicle is the responsibility of the user or integrator." It adds that "anyone who can send MAVLink to the vehicle over an unsigned link can command it, change its parameters and upload firmware. That is the design, not a vulnerability." Vulnerabilities, in ArduPilot's terms, are issues that defeat a protection it documents: MAVLink2 signing, secure firmware with a signed bootloader, and Remote ID tamper protection.

PX4's SECURITY_SCOPE.md takes the same approach. "PX4 protects one thing: who gets to fly/control the aircraft, and who gets to make it stop." As shipped, "every control interface is unauthenticated, unsigned and unencrypted." The SD card, USB, the onboard shell and the peripheral buses are trusted as the operator. The document also notes that the offboard transports used by companion computers, uXRCE-DDS and Zenoh, reach the flight controller's internal message bus directly, so "the command guards that apply to MAVLink do not apply to them."

Diagram of the open source autopilot security boundary: the telemetry radio, RC link and companion network above the flight controller are the integrator's to secure; the SD card, USB and shell, and peripheral buses below it are trusted as the operator

Figure 5. PX4 and ArduPilot put the radio link and companion network on the integrator's side of the line.

That is a reasonable engineering position, and it is honest. It also means that a drone built on either autopilot is only as secure as the integrator made it. If the integrator shipped with signing off and a plain radio, the aircraft accepts commands from anyone on the link.

PX4's published advisories

PX4 publishes advisories on its GitHub security page. A selection:

AdvisoryCVEWhat it wasFixed in
GHSA-5hvv-q2r5-rppwCVE-2023-46256Heap buffer overflow in a laser rangefinder parserAffected 1.14.0-rc1 and earlier
GHSA-qpw7-65ww-wj82CVE-2023-47625Global buffer overflow in the CRSF RC packet parser1.14.0
GHSA-mqgj-hh4g-fg5pCVE-2026-32706Oversized CRSF packet overflows a fixed 64-byte buffer1.17.0-rc2
GHSA-fh32-qxj9-x32fCVE-2026-32709Unauthenticated path traversal in MAVLink FTP: read, write, create, delete and rename files on the flight controller1.17.0-rc2
(listed on the advisory page)CVE-2026-32724Use-after-free in the MAVLink shell from a race between threads1.17.0-rc1
(listed on the advisory page)CVE-2026-32743Stack buffer overflow in the MAVLink log handlerAffected 1.17.0-rc2 and earlier
GHSA-mq93-4g5w-w34hnoneCode injection in a GitHub Actions documentation workflow running in a privileged contextFixed on main

Three things are worth drawing out.

The parser bugs cluster around radio-facing code. RC protocol parsers, MAVLink handlers and sensor drivers are where malformed input arrives. They are also where fuzzing pays off most.

CVE-2026-32709 shows why the link boundary matters. The advisory describes an unauthenticated path traversal that let any MAVLink peer reach arbitrary files on the flight controller. On a signed or encrypted link, "any MAVLink peer" is a short list. On an open link, it is anyone in range.

One advisory was about the build pipeline, not the drone. GHSA-mq93-4g5w-w34h describes a GitHub Actions workflow that evaluated user-controlled content in a privileged context, which "could execute arbitrary code, potentially compromising the CI/CD pipeline." For an autopilot project, continuous integration (CI) is part of the supply chain that builds the firmware every downstream maker uses.

How fixes reach drones

PX4's security policy says that only the 1.18.x series currently receives security updates, and that fixes are developed as ordinary public pull requests. It explains the choice: "downstream users do not update quickly enough for a coordinated release date to protect them, and many are not tracking upstream closely at all." That sentence describes the core supply chain problem in drones. A fix exists upstream. Whether it reaches a given aircraft depends on whether the maker of that aircraft tracks upstream, rebuilds and ships an update, and whether the operator installs it.

Protections the projects ship

Both autopilots give integrators real controls:

  • MAVLink 2 signing, covered above.
  • Signed firmware and secure boot. ArduPilot documents tamper-proof firmware using signed bootloaders and signed firmware, so that "only firmware signed with one of the public-private key pairs will run on the autopilot." PX4 supports secure boot and its scope document warns that "the in-tree secure boot variant ships a publicly committed test key that an integrator must replace."
  • Parameter lockdown and Remote ID tamper protection in ArduPilot, listed on its security page.
  • Isolation of companion transports, which PX4 calls out as essential because DDS and Zenoh bypass MAVLink's controls.

How do GNSS spoofing and jamming affect drones?

GNSS jamming drowns weak satellite signals so the drone loses its position fix; well-configured autopilots then hold, land or return using other sensors. GNSS spoofing feeds counterfeit signals that make the drone believe it is somewhere else, which is harder to detect because the fix looks healthy. Defences combine multi-constellation receivers, signal authentication, sensor cross-checks and failsafes.

Satellite navigation signals are extremely weak by the time they reach the ground. That is why they are easy to disrupt.

Jamming is the blunt version: noise on GNSS frequencies that stops the receiver from hearing the satellites. The drone loses its fix. A well-configured autopilot notices immediately and changes behaviour.

Spoofing is the subtle version. The Wikipedia article on GNSS spoofing describes an attack that broadcasts fake signals "structured to resemble a set of normal GNSS signals," or rebroadcasts genuine signals captured elsewhere or at a different time. A careful spoofer starts by matching the real signals, then increases power and slowly drags the receiver's computed position away. The receiver reports a healthy fix. It is simply wrong.

Two panels comparing GNSS jamming (receiver loses its fix; pilot sees a GPS lost warning; typical safe outcome is a failsafe) with GNSS spoofing (counterfeit signals pull the position away; pilot often sees nothing unusual; autopilot trusts sensors that report within spec), above five layered defences: multi constellation and multi band receivers, signal authentication such as Galileo OSNMA, sensor cross checks, failsafes, and planning and logging

Figure 6. Jamming removes the position. Spoofing replaces it with a false one.

A civil demonstration

The best-known civil demonstration is maritime rather than aerial, but the physics is the same. In 2013 a team from the University of Texas at Austin, led by Todd Humphreys, used a custom device to spoof the GPS receivers of an 80 million dollar private yacht and coerce it off course, after the team was invited aboard, to measure how hard spoofing was at sea and whether bridge sensors would catch it, as the university reported.

What the autopilot assumes

PX4's scope document is candid: "PX4 cannot reliably distinguish a real GPS fix, rangefinder return or gyro reading from a spoofed one. Some spoofing detection exists and is best-effort." It also says that a vehicle acting on a false position is not a vulnerability in PX4, while a spoofed signal that makes the receiver path corrupt memory would be. In other words, GNSS integrity is a system problem, not something firmware alone can solve.

Safety systems that exist

This is where drones are better placed than the headlines suggest. Autopilots have layers of fallback behaviour designed for exactly this kind of failure, because GNSS dropouts happen naturally near buildings, under bridges and close to metal structures.

ArduPilot Copter documents failsafes for radio loss, battery, ground station loss, EKF (the extended Kalman filter, the estimator that fuses sensor data into a position), dead reckoning, vibration and terrain data loss, plus crash checks, parachute support and an independent watchdog. PX4's safety configuration covers manual control loss, data link loss and position loss, with actions such as hold, return or land. Most consumer drones also fall back to an attitude mode, where the pilot flies manually without position hold.

These do not stop a spoofer, but they turn many interference events into a controlled landing rather than a flyaway.

Defences against spoofing

  • Multi-constellation, multi-band receivers. Wikipedia's spoofing article notes that using more than one navigation system at once is recommended to reduce exposure. A spoofer has to fake several constellations and frequencies consistently.
  • Signal authentication. Galileo's Open Service Navigation Message Authentication (OSNMA) signs navigation data using the TESLA scheme, is free to use, and has been operational since 24 July 2025, per the same article. Receivers that support it can reject forged navigation data. Our post on whether Galileo OSNMA stops GPS spoofing goes deeper into what it does and does not cover.
  • Sensor cross-checks. Compare GNSS against inertial dead reckoning, optical flow, visual odometry, barometric altitude and compass heading. Sudden jumps or a slow, steady drift that the IMU does not feel are red flags.
  • Failsafes configured on purpose. Decide in advance what the aircraft should do when position is lost or suspect, and test it in simulation.
  • Planning and logging. Check for known interference before operations, and review logs after any unexplained navigation event.

What data does a drone expose?

A drone's most sensitive output is usually not its video but its flight log, which records the full path, the home point and where every photo was taken. Logs, imagery and live views can leak through cloud sync, account takeover, shared files and lost hardware, and broadcast IDs reveal position to anyone nearby.

Table of drone data types: flight logs, photos and video, live view in fleet tools, broadcast Remote ID, proprietary broadcast IDs and account data, with what each reveals, where it lives and who can read it

Figure 7. What drone data reveals, where it lives and who can read it.

Flight logs

Every drone records telemetry. Check Point's research spelled out what that means: flight logs show the drone's exact location throughout the flight, and parsing them can reveal where each photo was taken. A pilot's logs over a few months are a map of their home, their workplace and their clients' sites.

Logs live in several places at once: on the aircraft, on the controller or phone, in the ground station's folders, in the vendor's cloud if sync is on, and in any third-party log analysis service the pilot uses. Each copy is a separate exposure.

Photos and video

Imagery is the point of most drone flights, and it can be sensitive in its own right: the inside of a construction site, a roof, a crop, a substation, a private garden. Many cameras also write GPS coordinates and time into file metadata, so a shared photo can carry the location it was taken from.

Live views and fleet tools

Enterprise fleet platforms show live camera and map views to authorised users. That is valuable, and it means that a fleet account is a real-time surveillance capability. Check Point's 2018 chain reached DJI's FlightHub live view for exactly this reason.

Lost and captured aircraft

Drones crash, run out of battery and land in places the pilot cannot reach. The SD card in a lost drone may hold the whole mission's imagery. PX4's scope document notes that the SD card also holds logs, the MAVLink signing key and sometimes boot scripts. A captured drone is a data breach and, if keys are stored on it, a credential compromise.

Vendor controls

Some vendors now offer settings to limit data leaving the device. DJI's Trust Center describes a Local Data Mode and cites an independent audit that "validated that Local Data Mode (LDM) resulted in no outbound traffic." Whatever the brand, look for an offline or local mode and use it for sensitive work.

What are the risks in drone companion apps?

Companion apps run with wide phone permissions, bundle third-party SDKs, sync data to the cloud and can command the aircraft. A 2020 analysis of the DJI GO 4 Android app found an update mechanism outside Google Play and an SDK collecting phone identifiers, and DJI firmware research showed most commands could come from a compromised phone.

The DJI GO 4 analysis (2020)

In July 2020 the security firm Synacktiv published an analysis of the Android version of DJI GO 4, the app used to fly many DJI drones at the time. Testing app version 4.3.36 with a Mavic 2, Synacktiv reported:

  • a self-update mechanism that could download and install updates or new applications outside Google Play, which "completely circumvents" Google's own update mechanisms;
  • a bundled third-party SDK (MobTech) that collected phone identifiers such as the IMSI and IMEI, which Synacktiv said were "not relevant or necessary for drone flights";
  • the app continuing to run and make network requests after the user swiped it closed.

Synacktiv also noted that the iOS version was not obfuscated and did not have the hidden update mechanism. A second firm, GRIMM, published a parallel report. According to Wikipedia's summary, DJI responded that the reports described "typical software concerns," and the researchers did not find evidence that data had been exploited.

The lessons are not specific to one vendor:

  • Permissions add up. A flight app legitimately needs location, storage, network and camera access. Any code running inside that app, including third-party SDKs, inherits all of it.
  • Update channels are trust channels. An app that can update itself outside the app store can change what it does without the store's review.
  • SDKs are suppliers. Analytics, crash reporting, social sharing and mapping SDKs each bring their own data collection and their own vulnerabilities.

The phone as a path to the aircraft

The NDSS 2023 researchers explicitly analysed an attacker who has compromised only the smartphone connected to the remote controller, with no physical access to the drone. Most of the commands they uncovered could be sent from that position over the air. So a malicious app, a vulnerable app, or a phone that has been compromised by unrelated malware is a route to the aircraft. Our guide on how phones get hacked covers how that happens.

Desktop ground stations

On the professional side, ground control software runs on laptops: Mission Planner as a Windows .msi, QGroundControl as a Windows installer, an AppImage on Linux and builds for other platforms. Those laptops also browse the web, read email and join hotel Wi-Fi. A GCS laptop that has signing keys, mission files and logs on it deserves the same protection as any other privileged workstation. See how laptops get hacked.

Defences

Install drone apps only from official sources and keep them updated. Fly from a dedicated phone or tablet where you can, and turn off permissions flying does not need. Enterprises should run apps under mobile device management with sideloading blocked. Makers should minimise permissions and SDKs, keep an SBOM (software bill of materials, the list of every component and version in a piece of software) for each app release, and ship updates only through the platform stores.

What does Remote ID mean for privacy?

Remote ID makes drones broadcast their identity, position, altitude and speed about once a second, and in the US and EU also the operator's or take-off position, in plain text that anyone with a phone app can read. Current standards leave authentication optional or unspecified, so broadcasts can be forged, and researchers are proposing privacy-preserving designs.

Remote ID is a digital licence plate for drones. Aviation regulators in the US, the EU, the UK and Japan now require it for most drones above small weight thresholds, and the industry standard behind most implementations is ASTM F3411, according to Wikipedia's Remote ID article. Broadcasts go out over Wi-Fi and Bluetooth Low Energy, and the article notes that "Remote ID information is meant to be broadcast to the public and readable by smart phones and similar devices."

What is broadcast

The Open Drone ID Core C library, an open source implementation used across the industry, documents what each rule requires. Its README shows that under both the US and EU rules the drone's dynamic position, timestamp, track direction and horizontal speed are mandatory, and the operator's dynamic position is mandatory too (an add-on module may use the take-off location instead). Location is sent at one-second intervals. An authentication signature is optional under ASTM F3411 and not required by the US or EU rules, though Japan's rule requires one.

So a Remote ID broadcast tells anyone nearby where the drone is, where it is going, and roughly where the pilot is standing.

The privacy problem

For a hobby pilot, the issue is personal safety and privacy: someone can walk to where you are standing. For a business, it is operational: a competitor or a curious member of the public can see where your inspection crews are working and when. Researchers have described this directly. A 2024 paper, Obfuscated Location Disclosure for Remote ID Enabled Drones, notes that Remote ID's plaintext broadcasts "allow malicious users to track drones via the disclosed information, possibly leading to drone capture and severe privacy leaks," and proposes broadcasting a differentially private location while giving authorised parties an encrypted exact one.

The authentication problem

Because most Remote ID broadcasts are not signed, they can also be forged or replayed. A 2025 paper, TBRD: TESLA Authenticated UAS Broadcast Remote ID, observes that current standards "leave authentication mechanisms underspecified, enabling spoofing, relay, and replay attacks," and proposes a lightweight authentication scheme compatible with existing standards. Anyone relying on Remote ID for situational awareness, such as a site operator watching for drones over a facility, should treat it as a useful signal, not proof.

DroneID is a separate thing

DJI's proprietary DroneID predates standard Remote ID and runs over DJI's own radio. The NDSS 2023 work showed it was readable with the right equipment. Pilots of DJI aircraft may therefore broadcast both.

What pilots and operators can do

  • Comply with Remote ID where the law requires it. It is a legal obligation, not an optional setting.
  • Know what your aircraft broadcasts, and plan where you stand with that in mind when it matters.
  • For enterprise crews, brief staff that their position is effectively public during flights.
  • For makers using ArduPilot, the project documents Remote ID support, and its security policy describes Remote ID tamper protection that refuses non-compliant firmware sent from a ground station.

How should you weigh these risks?

For most pilots and fleets, the biggest practical risks are account takeover and cloud data exposure, unsigned command links, and weak Wi-Fi on small drones, because they are cheap to exploit. Firmware exploits and GNSS spoofing are serious but take far more effort. Fix the cheap, high-impact items first.

Risk matrix plotting drone attack classes by attacker effort and impact: account takeover and synced logs, unsigned MAVLink links and open Wi-Fi are low effort with high impact; leaked keys, tampered app or update channels, firmware exploits and GNSS spoofing are higher effort with high impact; reading Remote ID and GNSS jamming sit lower; debug port extraction on a captured drone is high effort and lower impact for most operators

Figure 8. Impact for a typical commercial operator against attacker effort. A qualitative judgement from the cases in this guide.

The ordering above is a judgement, not a measurement, and it shifts with context. An agricultural operator flying over private farmland faces different threats from a utility inspecting a substation or a film crew in a city. But two rules hold almost everywhere. Data attacks are cheaper and more common than flight attacks. And the controls that stop the cheap attacks, such as account security, signed links and current software, are also cheap.

How do you detect attacks on drones?

Detection depends on who you are. Pilots watch for unexpected alerts, sign-ins and navigation behaviour. Fleets collect and review flight logs, account audit trails and network traffic from ground stations. Makers monitor vulnerability advisories for every component they ship and fuzz their own parsers.

Signals for pilots

Watch for sign-in alerts or sessions you do not recognise, the app disconnecting while the drone keeps responding to something, parameter or home point changes you did not make, GNSS warnings or position jumps, and update prompts from anywhere other than the official app.

Signals for fleets and operators

  • Flight logs reviewed centrally. Look for unexpected parameter changes, unknown ground station system IDs, failsafe triggers, GNSS anomalies and missions that do not match the work order.
  • Account and platform audit logs. New devices, new API tokens, unusual export volumes, logins from unexpected places.
  • Ground station network monitoring. Unexpected outbound connections from GCS tablets and laptops, and traffic to services you have not approved.
  • Firmware and app version drift. Aircraft or tablets running versions you did not deploy.
  • Physical events. Lost aircraft, lost SD cards and lost controllers are security events, not just insurance claims.

Signals for makers

  • New advisories in PX4, ArduPilot, the RTOS, chip vendor SDKs, video and networking libraries, and mobile SDKs you ship.
  • Crash reports from the field that point at parsers.
  • Fuzzing results from continuous runs against radio-facing code, an approach PX4's scope document describes through sanitiser builds of its simulator.
  • Secrets that appear in repositories, build logs or published packages.

Why is drone security a supply chain problem?

Much of the code in a drone was written by someone other than the drone maker: chip SDKs, an RTOS, an open source autopilot, generated MAVLink libraries, video and networking libraries, and mobile SDKs. A vulnerability in any of them flows into every product that uses it, and the fix has to travel back through every hop.

Flow diagram of the drone software supply chain: chip SDKs and HALs, RTOS, open source autopilots, MAVLink, open source libraries and mobile SDKs flow into the drone maker's flight firmware, companion image, ground control app and cloud services, built and signed in CI, then out through firmware updates and app stores to operators and fleets

Figure 9. Where drone software comes from, and why fixes take longer than vulnerabilities to travel.

The chain

Take a typical commercial drone built on an open autopilot:

  1. A chip vendor supplies the microcontroller, its hardware abstraction layer and often radio and video SDKs (see our post on fabless chip designers).
  2. An RTOS such as NuttX or ChibiOS runs under the autopilot, and PX4 or ArduPilot provides the flight stack.
  3. MAVLink message definitions are turned into code by a generator, and that code is compiled into the autopilot, the ground station and the companion software.
  4. The companion computer runs Linux, ROS 2, camera stacks and sometimes AI models (see industrial robotics security).
  5. The ground station app pulls in UI frameworks, map SDKs, analytics and crash reporting, and the cloud backend has its own dependency tree.
  6. A CI pipeline builds, tests and signs all of it.

A consumer drone from a large vendor has a different shape, with more in-house code, but the same pattern: third-party radio and video components, an embedded Linux somewhere, a mobile app full of SDKs, and a cloud backend.

Why fixes are slow

PX4's own security policy describes the problem: downstream users "do not update quickly enough" and many "are not tracking upstream closely at all." A drone maker that forked PX4 two years ago and has customised it since may not be able to pull in a fix without a significant merge. A maker that ships a vendor's radio module in binary form depends entirely on that vendor's release schedule. And operators often delay firmware updates during a busy season because a new version needs re-validation before commercial flights.

Secure updates are a supply chain control

Firmware signing, secure boot with the maker's own keys, and an update path that checks signatures are what make updates trustworthy. The NDSS 2023 signature bypass is a reminder to cover everything the bootloader loads, not just the main image. And the leaked keys in this guide, from DJI's 2017 SSL key to PX4's warning about its public test key, are reminders that signing is only as strong as key handling.

Regulation is moving

Drone makers selling in Europe will increasingly need to show this work. The EU Cyber Resilience Act requires manufacturers of products with digital elements sold in the EU to draw up an SBOM, and connected drones and their apps are products with digital elements. Our explainer on EU CRA SBOM obligations covers the deadlines and what the regulation asks for.

How should hobby pilots secure their drones?

Keep the aircraft, controller and app updated from official sources, protect the account with a unique password and two-factor authentication, turn off cloud sync you do not need, strip location data before sharing, set a Wi-Fi password where the drone uses Wi-Fi, configure failsafes, and know what your Remote ID broadcasts.

  1. Update everything from official sources. Aircraft firmware, controller firmware, batteries where they have firmware, and the app. Most of the vulnerabilities in this guide were fixed by an update.
  2. Secure your account. A unique password and two-factor authentication where the vendor offers it. Review active sessions from time to time.
  3. Decide what goes to the cloud. If you do not need flight log or media sync, turn it off. Use an offline or local data mode for flights you would not want in someone else's database.
  4. Strip metadata before sharing. Remove GPS data from photos and video you post publicly, especially ones taken near home.
  5. Lock down Wi-Fi drones. Set a strong network password if your drone creates a Wi-Fi network, and do not fly drones that cannot be secured near people or property.
  6. Protect the phone or tablet. Screen lock, current OS, apps only from official stores. Consider a dedicated device for flying.
  7. Configure failsafes. Return to home altitude, battery thresholds and signal loss behaviour, set for where you fly.
  8. Know your broadcasts. Comply with Remote ID, and understand that your position is visible to people nearby.
  9. If you build your own drone on PX4 or ArduPilot, enable MAVLink signing over USB before first flight, keep firmware current, and do not expose a MAVLink port to the internet.

How should businesses secure a drone fleet?

Treat the drone programme like any other operational technology: inventory every aircraft, controller, tablet, app and firmware version; segment it with zones and conduits; use signed or encrypted links; control accounts with SSO and MFA; keep data in your own storage; secure remote access; monitor logs; back up configurations; and rehearse incident response.

Drones are now routine tools for inspecting power lines, pipelines, solar farms, rail, bridges, telecom towers, mines and farms. For operators of that infrastructure, a drone programme touches the same environments as their control systems. The practices below are written for commercial operators and apply equally to service providers who fly on their behalf.

Start with an asset inventory

You cannot secure what you have not listed. For each aircraft, record the model, serial number, flight controller firmware version, radio and Remote ID module versions, the controller and its firmware, the tablet or laptop and its OS, the app or GCS version, and which accounts and cloud services it uses. Keep the SBOMs your vendors supply against each entry. This inventory is what lets you answer "are we affected?" the day an advisory lands.

Segment with zones and conduits

The Purdue model describes industrial networks as levels, from the physical process at level 0 through sensors, control systems and up to enterprise IT, as summarised on Wikipedia. The IEC 62443 series adds the idea of zones and conduits: zones group assets with common security requirements, and conduits are the controlled paths between zones. Our explainer on the Purdue model and where it falls short covers the history.

For a drone programme that maps naturally to the layout in Figure 10.

Zones and conduits diagram for an enterprise drone fleet: a field zone with aircraft, controllers and docks; a flight operations zone with managed GCS tablets, a local fleet server and key and update staging; a data processing zone for log and imagery ingest; an enterprise IT zone; cloud and vendor services reached by allowlisted egress; and a site OT network with no direct path to drone systems

Figure 10. Zones and conduits for an enterprise drone fleet. Drone systems never join the plant's control network.

  • Field zone: aircraft, controllers and docks. The conduit to flight operations is the command link, which should be signed, encrypted, or both.
  • Flight operations zone: managed GCS tablets and laptops, any local fleet server, and staging for keys and updates.
  • Data processing zone: log and imagery ingest, mapping and inspection tools, retention and deletion. Imagery moves in through a brokered transfer, not a shared drive mounted everywhere.
  • Enterprise IT zone: identity, MFA and storage, reached from flight operations only through secure remote access.
  • Cloud and vendor services: fleet SaaS and updates, reached through allowlisted egress.
  • The site's OT network, if there is one: drone systems do not join it. If imagery or data has to reach plant systems, it goes through a conduit designed for that purpose.

Secure the links and the ground stations

Turn on MAVLink signing for open autopilots, provisioned over USB. Where you need confidentiality, run links over encrypted radios or a VPN. Issue managed tablets and laptops, enrolled in device management, with sideloading blocked and only approved apps installed. Lock screens, encrypt storage and keep the OS current.

Control accounts and remote access

Put fleet platforms behind single sign-on with MFA. Give pilots, data analysts and administrators different roles. Remove access promptly when people leave. For remote support of docks or field systems, use a secure remote access service with MFA and session logging, not a shared password or an open port.

Keep data where you can govern it

Decide where logs and imagery live and for how long. Use local or offline data modes where vendors offer them for sensitive sites. Encrypt removable media. Apply retention and deletion rules, because a log you deleted on schedule cannot leak.

Monitor

Collect flight logs centrally and review them for anomalies: unexpected parameter changes, unknown ground stations, GNSS events, failsafes. Feed fleet platform audit logs into your security monitoring. Watch outbound traffic from the flight operations zone.

Back up and plan for failure

Back up aircraft configurations, parameter files, mission plans and signing keys (in a secure store, separate from the aircraft). Keep a known-good firmware image for each model you fly.

Plan for the failures that are not attacks too. Train pilots on manual flight in attitude mode, configure failsafes, keep visual line of sight where rules require it, and brief crews on what to do when GNSS degrades. These safety systems are the reason most interference events end with a controlled landing.

Rehearse incident response

Write down what happens when an aircraft is lost, when an account is compromised, when a tablet is stolen, and when a vendor publishes an advisory for something you fly. Who grounds the fleet, who rotates keys, who checks logs, who talks to the client whose site was being inspected. Run the drill once a year.

How should drone makers build secure products?

Sign firmware and enable secure boot with your own keys, ship MAVLink signing on by default or use an encrypted link, fuzz every parser that faces a radio, keep SBOMs for firmware, apps and companion images, scan for secrets, track upstream advisories, minimise app permissions and SDKs, and publish a disclosure policy.

This section is for drone companies, integrators who build on PX4 or ArduPilot, and makers who sell kits or modules.

Aircraft and firmware

  • Secure boot and signed firmware with your own keys. Replace any test keys, and keep signing keys in hardware or a managed signing service, never in a repository.
  • Cover everything the bootloader loads. Configuration files, patches and secondary images need integrity checks too, as NDSS 2023 showed.
  • Disable or lock debug interfaces on production units, and set a console password if your RTOS supports one. PX4's scope document notes that NuttX supports a console password and that no PX4 board enables it.
  • Fuzz radio-facing parsers continuously. RC protocols, MAVLink handlers, sensor drivers and your own command protocol. Run sanitiser builds in CI.

Links

  • Ship MAVLink signing on by default, with keys provisioned at the factory or on first USB connection. Do not rely on the customer to turn it on.
  • Encrypt where confidentiality matters, with a radio or network layer that uses standard, reviewed cryptography.
  • Isolate companion transports. If you run DDS or Zenoh between the companion computer and the flight controller, keep that network private to the aircraft.

Apps and cloud

Every SDK is a supplier with access to your users' phones, so keep the list short. Update only through the platform stores and your signed firmware path. Isolate authentication across services, protect session tokens, give users clear controls over what syncs, and offer a local data mode for customers who need one.

Supply chain and disclosure

  • Keep an SBOM for every release of firmware, companion images, apps and cloud services. Ask your suppliers for theirs.
  • Track upstream advisories for PX4, ArduPilot, your RTOS, chip SDKs and every library you ship, and keep your fork close enough to upstream that you can take a fix.
  • Scan every repository for secrets. A leaked signing key or cloud credential turns a single mistake into a fleet-wide problem.
  • Harden your CI. Treat build workflows as production systems, as PX4's workflow advisory shows.
  • Publish a vulnerability disclosure policy and a security contact, and fix fast. DJI, PX4 and ArduPilot all publish one.

Three hardening checklists: hobby pilots (updates, account security, cloud sync, local data mode, metadata, Wi-Fi password, failsafes, Remote ID), enterprise fleets (inventory, managed GCS devices, signed or encrypted links, zones and conduits, own storage, SSO and MFA, monitoring and GNSS planning, incident response) and makers (secure boot, MAVLink signing by default, fuzzing, SBOMs, secrets scanning, upstream advisories, minimal permissions and SDKs, disclosure policy)

Figure 11. Hardening checklists for the people who fly, operate and build drones.

How Safeguard helps teams that build drone software

Safeguard covers the software supply chain inside a drone product in one platform: SBOMs for firmware, apps and companion images, binary and image scanning, vulnerabilities ranked by reachability, SCA, SAST and secrets scanning on code you own, fix pull requests gated by your CI, Gold patched packages, supplier risk and compliance evidence.

Safeguard does not sit on the radio link or inside the flight controller's control loop. It sits where most of the long-lived exposure in this guide comes from: the firmware forks, companion images, open source components, apps and cloud services that drone makers and integrators build and maintain. Instead of one tool for source scanning, another for binaries, a spreadsheet of SBOMs and a separate GRC tool, Safeguard does that work end to end with one inventory and one vulnerability model, from the first SBOM to the merged fix and the audit record.

Flow diagram of Safeguard for drone products: source repositories, app binaries, images, extracted firmware and supplier SBOMs feed SBOMs, reachability ranking, SCA, SAST and secrets scanning, and policies and gates, leading to fix pull requests, Gold patched packages, supplier risk, compliance evidence and SBOM delivery

Figure 12. How Safeguard fits a drone product, from inputs to fixes and evidence.

See inside the binaries and images you ship and receive

In the Safeguard console, Integrations > Upload App Binary accepts compiled software: Windows EXE, MSI and DLL files, Android APKs and iOS IPAs, VM and disk images in QCOW2, VMDK, VHD and OVA formats, and container tarballs. Safeguard unpacks the file, identifies embedded libraries and their versions, maps them to known vulnerabilities and records licences, and the result becomes a project with a full SBOM. For drone teams that covers the flight apps on both mobile platforms, desktop ground stations and fleet tools, Linux companion computer images, and the containers behind fleet backends. Raw flight controller and radio firmware images may need to be extracted into their component files first, and the usual limits of binary analysis apply: heavy obfuscation, encryption or unusual packing can give incomplete results. Our guide to scanning app binaries with Safeguard and our explainer on binary composition analysis go further.

Rank vulnerabilities by what can actually be reached

A companion image or app can contain hundreds of components, and many vulnerabilities in them sit in code your product never calls. Safeguard's reachability analysis ranks the vulnerabilities your code can actually reach above those it cannot, and adds EPSS (the Exploit Prediction Scoring System) and known exploited context. That is how a team tells the parser bug on a radio-facing path that needs fixing this week from an unused function in a bundled library.

Scan the code you own

For your autopilot fork, companion software, apps and backend, Safeguard runs software composition analysis (SCA) across the full transitive dependency tree, static application security testing (SAST) on your own code, and secrets scanning that catches signing keys, API tokens and cloud credentials before they ship. The classes of bug in this guide, buffer overflows in parsers, path traversal in file handling, injection in CI workflows and credentials published to a repository, are what those scanners exist to catch. Our secrets scanning guide covers the practice.

SBOMs your customers and auditors can use

Every project produces an SBOM you can export in CycloneDX or SPDX, among other formats. That is what an enterprise fleet buyer increasingly asks for, what a regulator will expect under the EU Cyber Resilience Act, and what lets you answer "do we ship the affected PX4 version, which MAVLink library, which kernel" in minutes. Safeguard's Third-Party Risk Manager lets you request SBOMs from your own suppliers, such as radio, camera and Remote ID module vendors, score them and monitor them continuously. See SBOMs for firmware and IoT devices and SBOMs for embedded systems.

Fix it without waiting for upstream

This is where Safeguard differs most from a scanner. For code in your repositories, Griffin, Safeguard's remediation model, opens a fix pull request that your own CI builds and tests before anyone merges it, and your policies decide what can merge. For open source components where upstream has not shipped a fix, or where the fix sits in a release you cannot move to without a large merge, Safeguard Gold provides patched, hardened builds of those packages, and Safeguard can fork and patch the vulnerable package so your next build pulls a fixed version. For a drone maker on a long-lived autopilot fork, or a companion image pinned to an older distribution, that is the difference between shipping the fix in the next update and waiting for a quarter's worth of upstream changes to be merged. Closed source components still have to come from their vendor, but the open source around them no longer has to wait.

Compliance evidence from the same data

Safeguard's compliance suite collects evidence for frameworks such as SOC 2 and ISO 27001 from the same scans, policies and fix history. For drone makers preparing for the EU Cyber Resilience Act, or answering enterprise and critical infrastructure customers' security questionnaires, the SBOMs, vulnerability monitoring records, risk decisions and fix trail are the raw material those reviews ask for.

Deploy where your code lives

Flight controller source, signing infrastructure and customer imagery pipelines are sensitive. Safeguard runs as SaaS, in your own VPC, or fully on premises. It can run in Zero AI mode with no language model in the loop, or with a model you choose.

How this compares

Device makers have other good options. Finite State describes itself as "The Autonomous Product Security OS for Connected Devices," and ONEKEY offers "Effortless Product Cybersecurity & Compliance" with firmware analysis and SBOM management. Both focus on connected device product security and firmware. Safeguard's case is breadth and remediation across the whole drone software estate: one inventory across source, app binaries, companion images and containers; one vulnerability model with reachability; fix pull requests and patched packages instead of tickets; and supplier risk and compliance evidence from the same data. Safeguard can also work alongside an existing firmware analysis tool and take over SBOM delivery and fixing.

The takeaway

Drone hacking has a cinematic reputation: a stranger seizing an aircraft out of the sky. That has been demonstrated, on open Wi-Fi drones in 2013 and in lab conditions on firmware since. But the attacks that matter most to real pilots and real fleets are quieter. An account cookie stolen through a forum. A signing key on GitHub. A telemetry link left unsigned because that was the default. A flight log that knows where you live. A broadcast that tells everyone nearby where the pilot is standing.

Nearly all of the documented cases were fixed, and the fixes share a shape. Someone found the flaw, told the company or project that wrote the software, and that company shipped an update that then had to travel through forks, builds, app stores and operators before it reached an aircraft. Pilots can shorten that journey by updating and securing their accounts. Fleets can shorten it by knowing exactly what they fly and segmenting it properly. And makers can shorten it most of all, by knowing what is inside every firmware image, companion computer and app they ship, and by being able to fix it without waiting for everyone upstream.

Try it yourself

Click through these interactive Safeguard demos to see the steps described in this guide.

Test yourself

Why can anyone in range take over a small drone that runs on open Wi-Fi?

The drone's network has no password and the drone accepts commands from whoever is connected.

How secure is MAVLink?

MAVLink 1 has no authentication, recorded as CVE-2020-10282. MAVLink 2 adds message signing with a shared 32-byte key and a timestamp to block replays, but it is off unless configured and it never encrypts the payload.

What did Check Point show about DJI accounts in 2018?

A web flaw could hand an attacker a DJI account along with its synced flight logs, photos and live fleet views. DJI fixed the reported issue.

What did the 2023 DroneID research show?

Researchers from Ruhr University Bochum and CISPA showed that the DroneID broadcast was not encrypted and could be read by anyone with inexpensive hardware, compromising the operator's privacy. They also found 16 firmware vulnerabilities.

What is the difference between GNSS jamming and spoofing for a drone?

Jamming drowns weak satellite signals so the drone loses its fix, and a well-configured autopilot then holds, lands or returns. Spoofing feeds counterfeit signals so the drone believes it is somewhere else, which is harder to detect because the fix looks healthy.

Why is a flight log often more sensitive than the video?

It records the full path, the home point and where every photo was taken. Logs can leak through cloud sync, account takeover, shared files and lost hardware.

Frequently asked questions

Can a drone be hacked?

Yes. Documented research has shown drone accounts taken over through web flaws, small Wi-Fi drones taken over in flight, firmware bugs that crash drones mid-air, and broadcast identity data read with inexpensive radios. Nearly all of these were fixed by vendor updates. Keeping firmware, controller and app current, and securing your account, removes most of the practical risk.

Can someone take control of my drone in flight?

It depends on the link. Small drones on open Wi-Fi have been taken over in flight. Open autopilots using MAVLink accept commands from anyone on the link unless message signing or an encrypted link is in place. Large consumer drones use proprietary encrypted links, and the best-documented flight attacks on them needed a compromised phone or a firmware bug that has since been fixed.

Is MAVLink secure?

MAVLink 1 has no authentication, recorded as CVE-2020-10282. MAVLink 2 adds message signing with a shared 32-byte key and a timestamp to block replays, but it is off unless configured, and it never encrypts the payload. For a secure link, provision signing over USB before flight and run the link over an encrypted radio or VPN where confidentiality matters.

Do drone apps collect my data?

Drone apps need location, storage and network access to work, and many sync flight logs and media to the cloud. A 2020 analysis of the DJI GO 4 Android app found an SDK collecting phone identifiers and an update path outside Google Play; DJI described the reports as typical software concerns. Review permissions, turn off sync you do not need, and use offline or local data modes for sensitive flights.

What does Remote ID broadcast about me?

Remote ID broadcasts the drone's identifier, position, altitude, direction and speed about once a second. Under US and EU rules it also includes the operator's position, or the take-off location for add-on modules. Anyone nearby with a phone app can read it, and most broadcasts are not authenticated, so they can also be forged.

Can GPS spoofing make a drone crash or fly away?

GPS spoofing can make a drone believe it is somewhere else, which could steer it off course, and autopilots cannot reliably tell a spoofed fix from a real one. Jamming is more common and usually triggers a failsafe such as hold, land or return. Multi-constellation receivers, Galileo OSNMA authentication, sensor cross-checks and tested failsafes reduce both risks.

How do I secure a drone fleet for my business?

Start with an inventory of every aircraft, controller, tablet, app and firmware version. Segment drone systems into zones with controlled conduits and keep them off plant control networks. Use signed or encrypted links, managed tablets, SSO with MFA, secure remote access, central log review, configuration backups and a rehearsed incident response plan.

What is a drone SBOM and why does it matter?

A drone SBOM is a software bill of materials listing every component and version in a drone's firmware, companion computer, apps and cloud services, such as the PX4 or ArduPilot version, the RTOS, MAVLink libraries and mobile SDKs. When an advisory lands, it tells a maker or fleet operator in minutes whether they are affected, and it is increasingly required by enterprise buyers and the EU Cyber Resilience Act.

Sources

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

We email you a link to confirm, and only then add you to the list. Unsubscribe in one click. Privacy

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.