Microsoft 365 vs Google Workspace vs the rest: a security comparison of office suites, by the numbers

Short answer. No office suite is "the secure one", and the numbers show it in an unexpected way. Microsoft 365 has by far the most published vulnerabilities of any suite (1,005 CVEs for its office products so far in 2026, by Microsoft's own data) and the most entries on CISA's list of exploited flaws (78 in total). That reflects how much installed and self-hosted software it ships, how many people attack it, and how openly Microsoft discloses. CVE counts measure disclosure practice and attack surface, not code quality. Google Workspace has almost no CVEs, because its apps are web services Google fixes on its own servers, but its users live inside Chrome, which has 74 exploited entries of its own. Zoho, Apple iWork, Proton and LibreOffice have very small public footprints. ONLYOFFICE picked up its first exploited entry in October 2026. The suite that fits you depends on where you want the patching burden to sit, which controls you need, where your data must live, and how you plan to govern AI assistants.

This week explains why the question matters. On 10 November 2026 Microsoft patched CVE-2026-124194, an Outlook flaw that was already being exploited to leak NTLMv2 hashes, and CISA added it to the Known Exploited Vulnerabilities catalog the same day. Two days later a large Microsoft 365 outage took down Exchange Online mailboxes and disrupted Teams, SharePoint and Copilot. Microsoft said there was no indication the outage was the result of malicious or cyber activity. The two events are not connected; we covered both in detail in our incident analysis. Together they raise two questions: how exposed is the suite we run, and what happens when it goes down?

The numbers at a glance

Sources and exact queries follow below. Where a number is not public, we say so rather than guess.

Microsoft's CVE figures come from Microsoft's own Security Update Guide, where Microsoft is the numbering authority for its products. The other suites' CVE figures come from the National Vulnerability Database (NVD). KEV counts come from CISA's catalog feed, version 2026.11.10. The Project Zero columns count rows in Google Project Zero's public "0day In the Wild" tracker; the 2026 tab is a snapshot, last updated with a patch dated 9 November 2026.

How we counted

Each vendor publishes differently and each database indexes differently, so here is exactly what we did.

Microsoft 365. We queried the Microsoft Security Update Guide API (api.msrc.microsoft.com, sug/v2.0, the affectedProduct endpoint) for the product families "Microsoft Office", "Server Software", "Apps", "Mobile" and "Cloud", filtered by each CVE's initial release date in each calendar year. We kept rows where Microsoft is the issuing numbering authority and the product name matches Office, Outlook, Word, Excel, PowerPoint, Access, OneNote, Visio, Publisher, Project, Exchange, SharePoint, Teams, OneDrive, Skype or "365". We then counted distinct CVE IDs. Third-party CVEs that Microsoft republishes, such as Chromium bugs in Edge, are excluded. For 2026, 770 of the CVEs touch the Office apps, 205 SharePoint, 52 Exchange and 16 Teams (one CVE can touch several).

Everyone else. We queried the NVD CVE API 2.0 with rejected records excluded, and bucketed results by the NVD "published" date. Keyword phrases used exact matching against the CVE description. The queries were: Google Workspace apps: "Google Workspace", "Gmail", "Google Docs", "Google Drive", "Google Meet". Chrome: "Google Chrome". Zoho Workplace: "Zoho Mail", "Zoho Writer", "Zoho Sheet", "Zoho Show", "Zoho WorkDrive", "Zoho Cliq", "Zoho Workplace". Apple: the CPE match strings cpe:2.3:a:apple:pages, apple:numbers, apple:keynote and apple:icloud, plus the keyword "iCloud". LibreOffice: keyword "LibreOffice" plus cpe:2.3:a:libreoffice:libreoffice. ONLYOFFICE: keyword "ONLYOFFICE" plus cpe:2.3:a:onlyoffice. Proton: "Proton Mail", "ProtonMail", "Proton Drive" plus cpe:2.3:a:proton. Results from the queries in each row were merged and de-duplicated by CVE ID. Keyword searches also catch other products that merely mention a suite, such as WordPress plugins that embed Google Docs or document converters built on LibreOffice, so we read every result outside Chrome by hand and kept only the vendor's own products. For ONLYOFFICE we counted Docs (Document Server), DocSpace, Workspace, Community Server and the desktop editors, and left out connector plugins for other platforms (including ONLYOFFICE's own WordPress and ownCloud connectors) and third-party products that embed ONLYOFFICE. The NVD counts agree closely with Microsoft's own data: the same method applied to Microsoft gives 126, 90, 211 and 995 CVEs for 2023 to 2026, against 131, 80, 214 and 1,005 from the Security Update Guide.

Exploited in the wild. We downloaded CISA's KEV JSON feed (catalog version 2026.11.10, 1,779 entries) and filtered on the vendorProject and product fields. For Microsoft 365 we counted the products Office, Outlook, Office Outlook, Word, Excel, PowerPoint, Publisher, Project, Office and WordPad, Visual Basic for Applications, Exchange, Exchange Server, SharePoint and SharePoint Server. For Chrome we counted Google entries naming Chrome or Chromium, plus the Chrome engine components Skia and Dawn. We counted by the date CISA added each entry, not by the CVE's own year.

Zero-days. We exported the per-year tabs of Google Project Zero's public 0day In the Wild spreadsheet as CSV and counted rows by its Vendor and Product columns.

How to read these numbers

Read the table as a map of disclosure and attack surface, not as a quality ranking. CVE counts measure how much a vendor discloses and how much software customers run and patch themselves. They do not measure how well the code is written, and a higher count is not evidence of a less careful vendor. Five caveats matter.

1. Cloud services often get fixed without a CVE. When a bug lives only on a vendor's servers, nobody has anything to install, so historically there was nothing to number. Microsoft changed its policy in June 2024: it said it "will issue CVEs for critical cloud service vulnerabilities, regardless of whether customers need to install a patch or to take other actions to protect themselves" (MSRC, 27 June 2024). In Microsoft's 2025 and 2026 data we count 8 and 31 office-suite CVEs where no customer action was required, most of them for SharePoint Online, Teams and Exchange Online. Google, Zoho and Proton run web-first services and have no matching public commitment that we could find, so a low count for them partly means "fixed quietly on the server".

2. Installed and self-hosted software accumulates CVEs. Anything customers must patch themselves gets a CVE. Microsoft ships Office for Windows, Mac, iOS and Android, several LTSC editions, and on-premises Exchange and SharePoint servers. LibreOffice and self-hosted ONLYOFFICE are installed software too. The cloud-heavy suites have far less of that surface.

3. Disclosure practice varies by vendor. Microsoft, Google, Apple, The Document Foundation and Proton AG are all CVE Numbering Authorities, according to the CVE program's partner list. Zoho's numbering authority scope covers its "ManageEngine, Zoho, and Zakya branded on-premise products only", so Zoho's cloud services sit outside it. ONLYOFFICE's developer is not listed. A vendor that numbers its own bugs tends to publish more of them.

4. Attention follows market share. Researchers and criminals both go where the users are. Microsoft paid $17 million in bounties in the year to June 2025 (MSRC), which buys a lot of found bugs.

5. The 2026 jump is real, and it is not only about quality. Microsoft's office-suite CVE count was 19 in January 2026 and 38 in May, then 109 in July, 221 in October and 217 in the first half of November. Microsoft's November 2026 Patch Tuesday alone listed more than 1,000 CVEs across all its products, a total that includes third-party CVEs Microsoft republishes, as we noted in our incident analysis. We don't attribute a cause, because none is published in the sources we read, and the jump is not unique to Microsoft. Chrome's NVD count went from 204 in 2025 to 3,975 so far in 2026, and LibreOffice's from 7 to 19. For planning, the point is simpler: the monthly volume your team must triage is now an order of magnitude higher than in 2024, whichever suite you run.

The KEV count is the most useful single column, because it measures confirmed exploitation, not disclosure volume. A zero in the KEV column means "no confirmed exploitation on record", not "unexploitable".

Suite by suite: what the numbers say

Microsoft 365

Microsoft 365 carries the largest and most exploited footprint of any suite here. Of its 78 KEV entries, 33 are Exchange or SharePoint, the on-premises servers many organizations still run beside Exchange Online and SharePoint Online. The 2025 and 2026 additions show the pattern clearly. The July 2025 SharePoint "ToolShell" chain (CVE-2025-53770, with CVE-2025-49704 and CVE-2025-49706) hit on-premises SharePoint servers; we wrote up the patch response at the time. In 2026, CISA added eight more SharePoint entries, an Exchange Server flaw, Office security-feature bypasses such as CVE-2026-21509, and this week's Outlook flaw.

Two details in the KEV list are worth knowing. First, CISA keeps adding very old Office bugs: CVE-2007-0671 (Excel, added August 2025), CVE-2009-0556 (PowerPoint, January 2026), CVE-2009-0238 (Office, April 2026) and CVE-2012-1854 (VBA, April 2026). Second, the Outlook credential-leak class keeps coming back: CVE-2023-23397, CVE-2024-21413 (added to KEV in February 2025), and now CVE-2026-124194.

Google Workspace

Our NVD queries found no CVEs for Google's own Workspace apps published from 2023 to 2026; every keyword match was a third-party plugin or tool that integrates with Gmail, Docs or Drive. There are no KEV entries either. Gmail, Docs, Drive and Meet run on Google's servers and update without customer action, so there is rarely anything to number. The trade-off is that the client is the browser. Chrome has 74 KEV entries, 7 added in 2025 and 8 so far in 2026, and Project Zero logged 6 to 8 exploited Chrome zero-days in each year from 2023 to 2026. If your Workspace users run Chrome, Chrome's patch level is effectively your Workspace client's patch level. Google Threat Intelligence Group counted 90 zero-days exploited in the wild in 2025, with browsers making up "less than 10%" of them.

Zoho Workplace

Zoho's cloud office products have no CVEs in our 2023 to 2026 queries. The three Zoho matches we found were for Zoho-made WordPress plugins (ZeptoMail and Zoho Mail), not for the Workplace services themselves. All nine Zoho entries in KEV are ManageEngine IT-management products, mostly from 2021 and 2022, which share a parent company with Zoho Workplace but not a code base or a deployment model. Zoho runs a public bug bounty that covers "All Zoho branded products and applications listed at zoho.com" (Zoho Bug Bounty).

Apple iWork and iCloud

The NVD holds two iWork CVEs published in our window, both on 28 January 2026: CVE-2025-46316 in Pages and CVE-2025-46306 in Keynote. CISA's catalog has no entry naming iWork or iCloud services. Apple's 96 KEV entries are almost all operating system and WebKit flaws. One 2025 entry, CVE-2025-43200, concerns "a maliciously crafted photo or video shared via an iCloud Link", but the flaw was in how Apple's operating systems processed the media, not in iCloud Drive or iWork. For iWork users, the operating system patch level is the number to watch.

LibreOffice

LibreOffice publishes its fixes on its own security advisories page, which lists 4 advisories with 2023 CVE IDs, 6 with 2024 IDs, 4 with 2025 IDs and 22 with 2026 IDs (a CVE's ID year can differ from its publication year, which is why these differ slightly from the NVD columns). It has no KEV entries and no rows in Project Zero's tracker. As installed desktop software, its exposure depends entirely on how fast you roll out new versions.

ONLYOFFICE

ONLYOFFICE is the one "small" suite with confirmed exploitation. On 8 October 2026 CISA added CVE-2021-3199, a path traversal in ONLYOFFICE Docs that "could allow for remote code execution" when JWT is used, via an image upload parameter. The fix shipped years earlier, in Document Server 5.6.3, according to the project changelog that CISA links. That is the self-hosting lesson in one entry: a five-year-old fix only protects servers that took it.

Proton

Our queries found no CVEs for Proton Mail or Proton Drive published from 2023 to 2026 (the few Proton entries in the NVD are for its VPN client on Windows), and nothing in KEV. Its strength is architectural: message bodies and attachments are stored with end-to-end encryption. Its documentation is candid about the limits too: "Subject lines in Proton Mail messages are not end-to-end encrypted", and mail to non-Proton addresses "isn't end-to-end encrypted by default" after delivery (Proton support).

Bug bounty and disclosure programmes

A bounty shows how much a vendor pays to hear about bugs before attackers use them.

Outages in 2025 and 2026

Availability is a security property too: when the suite is down, alerts, approvals and password resets don't arrive. These are the major incidents we could confirm from a vendor page or mainstream reporting.

For Zoho Workplace, Apple iCloud and ONLYOFFICE's cloud we found no 2025 or 2026 incident with a vendor report or mainstream coverage we could verify, so we list none. LibreOffice is installed software, so it has no service to go down.

Admin security controls

The control sets differ more by licence tier than by vendor, so check the edition you actually pay for.

The sources behind each cell: Microsoft's Conditional Access licensing ("Using this feature requires Microsoft Entra ID P1 licenses", with Business Premium also included, and "Security defaults ... are available for all customers"), Purview DLP and eDiscovery; Google's 2-step verification, context-aware access, DLP, Vault and data regions pages; Zoho's data centre page; and ONLYOFFICE's security page. "Not published" means we did not find it in the vendor pages we read, not that the feature is missing.

Microsoft 365 has the deepest control set here, but its strongest controls sit in higher tiers (Entra ID P1 or P2, E5), and Google's context-aware access and Vault are missing from its entry tiers. Compare prices at the tier that has the controls you need.

Default settings that matter

Defaults decide what happens in organizations that never change them.

Microsoft 365. Three defaults close old attack paths. Outlook "is configured by default to block automatic picture downloads from the Internet" (Microsoft Support), which stops tracking pixels and some forced-connection tricks. Since 2022 Office blocks VBA macros in files from the internet: Microsoft changed "the default behavior of Office applications to block macros in files from the internet" (Microsoft Learn). That change covers Access, Excel, PowerPoint, Project, Publisher, Visio and Word on Windows, and per the same page it "doesn't affect Office on a Mac, Office on Android or iOS devices, or Office on the web". Exchange Online has also turned off legacy sign-in: "Basic authentication is now disabled in all tenants" (Microsoft Learn).

NTLM is the default that hasn't changed yet. It is the protocol this week's Outlook flaw abused, and Windows still supports it. Microsoft's January 2026 plan, "Disabling NTLM by default", makes enhanced NTLM auditing available now, fixes common blockers in the second half of 2026, and disables network NTLM by default only "in the next major Windows Server release and associated Windows client releases". Until then, blocking outbound NTLM is your job, not a default.

Google Workspace. Google's 2-step verification can be made optional or required, and Google states it is "enforcing 2SV for administrator accounts" (Google Workspace Admin Help). The same page calls security keys "the most secure form of 2SV".

LibreOffice. LibreOffice has four macro security levels. At High, "only macros from trusted sources and signed macros (from any source) are allowed to run"; at Very high, "only macros from trusted file locations are allowed to run" (LibreOffice Help).

ONLYOFFICE. Macros are JavaScript: "Recorded macros are generated as JavaScript code that relies on Docs API" (ONLYOFFICE Help Center). They run inside the editor through its API rather than as VBA, so VBA-specific macro guidance does not carry over; review ONLYOFFICE's own macro settings for your deployment.

Patching model: cloud versus installed clients

Where the patching burden sits is the biggest structural difference between these suites.

Microsoft 365 splits the work. Microsoft patches Exchange Online, SharePoint Online and Teams services itself. Customers patch everything installed: Microsoft 365 Apps, Office LTSC, Outlook for Mac and mobile, and any on-premises Exchange or SharePoint servers. Microsoft 365 Apps ship through update channels (Microsoft Learn): Current Channel is the default and gets features "as soon as they're ready (once a month)"; Monthly Enterprise Channel updates "once a month, on the second Tuesday of the month"; Semi-Annual Enterprise Channel takes features twice a year but security fixes monthly, and "beginning July 2026, Semi-Annual Enterprise Channel will begin receiving monthly feature and security updates". Office LTSC uses its own perpetual channel. The practical risk is deferral: rings that hold updates back for weeks are exactly where an exploited Outlook flaw lives longest.

Google Workspace is patched by Google, server side, with nothing for customers to install for the apps themselves. The customer's patching job moves to the browser and the device: Chrome versions, ChromeOS, and mobile apps.

Zoho Workplace and Proton follow the same cloud model. Apple iWork updates through the App Store, and its security largely rides on the operating system's patch level.

LibreOffice and self-hosted ONLYOFFICE put the whole job on you. LibreOffice's advisories list which versions fix each issue, and nothing changes on a desktop until someone installs the new build. Self-hosted ONLYOFFICE Docs and DocSpace servers are internet-facing web applications, and CVE-2021-3199 shows what happens to the ones nobody updates.

Compliance attestations and data residency

Every cloud suite here publishes attestations, but they cover different things, and none of them transfers to your own audit automatically. You still have to show how you configured the service.

Microsoft 365. "Microsoft commissions a full SOC 1 Type 2 and SOC 2 Type 2 examination of Office 365 annually" (Microsoft Learn). Microsoft 365 holds an ISO 27001:2022 certificate, with a combined statement of applicability covering ISO 27001, 27017, 27018 and 27701 (Microsoft Learn). For residency, the EU Data Boundary is a "geographically defined boundary within which Microsoft has committed to store and process Customer Data and personal data" for the EU and EFTA, with "limited circumstances" where data still leaves it. Customers who bought Multi-Geo are not in scope for the boundary.

Google Workspace. Google lists ISO/IEC 27001, 27017, 27018 and 27701 among "the key international standards we are audited against", along with "SOC 2 and SOC 3 reports" (Google Workspace security whitepaper). Data regions let supported editions keep covered data in the United States or Europe.

Zoho Workplace. Zoho states it "has earned ISO/IEC 27001 certification" (valid to August 2028), lists ISO 27017, 27018 and 27701, and says "Zoho is SOC 2 Type 2 compliant" (Zoho compliance). It offers the widest choice of regions here: the United States, Europe, India, Australia, Japan, Canada, Saudi Arabia, the UAE and Singapore, with China listed separately (Zoho data centres).

Apple iWork and iCloud. Apple's ISO/IEC 27001 and 27018 certifications cover iCloud among other services (Apple certifications). A SOC 2 report for iCloud: not published.

LibreOffice. Attestations don't apply in the same way, because there is no service. Files stay wherever you store them, so your own storage's controls and attestations are the ones that count.

ONLYOFFICE. The vendor's security page says "ONLYOFFICE itself is not currently ISO-certified as a standalone product", states that it is compliant with GDPR, and says data in its cloud "is stored in certified data centers" (ONLYOFFICE security). Self-hosted, it inherits the attestations of whatever infrastructure you run it on.

Proton. Proton reports its "first SOC 2 Type II attestation" in July 2025 and ISO 27001 certification in May 2024 (Proton; Proton trust page). Proton AG is headquartered in Geneva.

For your own audits, the useful question isn't "is the vendor certified?" It's "can we prove how we configured it, and that we checked it again this quarter?" That evidence is where most teams struggle.

AI assistants and how they handle your data

Every major suite now ships an assistant, and every vendor says it won't train its models on your business data. The real differences are in where processing happens, what the assistant can see, and how well it resists prompt injection.

Microsoft 365 Copilot. Microsoft states that "Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs", and that Copilot "only surfaces organizational data to which individual users have at least view permissions" (Microsoft Learn). EU traffic "stays within the EU Data Boundary", with an exception the same page lists for some third-party models. Microsoft also notes that its jailbreak and cross-prompt injection classifiers "may not be available in all Microsoft Copilot scenarios". Copilot has already had a critical bug: CVE-2025-32711, an information disclosure flaw rated 9.3 under CVSS 3.1, which we analysed in our EchoLeak write-up.

Gemini in Google Workspace. Google says Workspace data is "not used for Generative AI model training outside your domain without permission", that chats "won't be reviewed by human reviewers", and that "Admins manage access to the Gemini app and can choose to disable the service" (Google Workspace privacy hub). Google describes a layered defense against indirect prompt injection in Gemini in Workspace apps, including content classifiers, markdown sanitization, suspicious URL redaction and user confirmation (Google; Workspace Admin Help).

Zoho Zia. "Zoho does not use customer data to train or improve its AI models", and by default "Zia uses Zoho's in-house LLM, built on open-source models and hosted within Zoho's infrastructure". Admins can connect an outside provider with their own key, in which case that provider's terms apply (Zoho).

Apple. Apple Intelligence runs on device or on Private Cloud Compute, where Apple says "no user data is retained in any form after the response is returned" and "User data is never available to Apple" (Apple Security Research).

ONLYOFFICE. Its AI agent lets you "link your own AI provider using the provider's API key or the locally hosted model" (ONLYOFFICE), so data handling follows the provider you pick, or stays local.

LibreOffice. We found no built-in AI assistant in LibreOffice's feature pages for its current release.

Proton. For its Lumo assistant, Proton says "we keep no logs of what you ask, or what Lumo replies", "Lumo doesn't use your chats to train the AI models", and chat history is synced "using zero-access encryption" (Proton). Its Scribe writing assistant "can be run locally", or on Proton's servers, and "does not train on your inbox data" (Proton).

The practical risk is less about training and more about reach. An assistant that can read everything a user can read will find the overshared salary spreadsheet in seconds, and a poisoned email or document can try to steer it. Fix permissions and sharing before a broad rollout, decide who gets the assistant, keep logs of prompts and responses where your suite allows, and test how the assistant behaves with untrusted content.

Decision guide: which suite fits which organization

There is no single winner. These are the fits the evidence suggests.

You run a Microsoft estate with Windows, Entra ID and Intune, or you need the deepest compliance tooling. Microsoft 365 fits, with the broadest control set here. Plan for the patch volume: retire on-premises Exchange and SharePoint where you can, keep update rings short, and license the tier that gives you Conditional Access.

You want the smallest patching burden and a browser-first workforce. Google Workspace fits. The apps patch themselves, so your job is managing Chrome and devices. Pick an edition with context-aware access and data regions if you need them.

You are a cost-conscious small or mid-sized business, or you need regional data centres. Zoho Workplace is worth a look: a small public vulnerability footprint, ISO 27001 and SOC 2 Type 2 attestations, and the widest choice of regions here.

You are an Apple-first team of designers, agencies or small companies. iWork and iCloud work well for documents but are not a full admin and compliance platform. Pair them with Apple Business Manager and keep the operating systems current.

You need full control of your data, offline use or open formats, or you have strict sovereignty rules. LibreOffice on the desktop, or self-hosted ONLYOFFICE for collaboration, keep files wherever you put them. The price is that patching, backups and hardening are all yours.

You handle sensitive communications: legal, journalism, health, advocacy, or executives. Proton's end-to-end encryption by default is the strongest privacy posture here. Remember the documented exceptions: subject lines, and mail exchanged with other providers.

Many organizations run a mix, which is fine if every suite is in your asset inventory and your vendor-risk register.

Fix it without waiting for upstream

These steps lower your exposure today, with the suite you already have.

Microsoft 365

1. Measure Office build versions across every endpoint against the latest security release for each channel, and shorten deferral rings for exploited fixes to days.
2. Block outbound TCP 445 to the internet on every Windows endpoint, and turn on outgoing NTLM auditing so you can see where NTLM goes before you restrict it.
3. Keep Outlook's default block on automatic picture downloads, and keep the internet-macro block on. For users who need no macros at all, Microsoft suggests "Disable all without notification".
4. Turn on security defaults or Conditional Access policies that require MFA for every account, and phishing-resistant methods for administrators.
5. If you still run Exchange Server or SharePoint Server, patch them first and restrict their internet exposure, since that is where most of Microsoft's KEV entries sit.
6. Review Outlook and Teams add-ins and third-party app consents, and remove anything nobody owns.

Google Workspace

1. Require 2-step verification for all users, and security keys or passkeys for administrators and high-risk staff.
2. Manage Chrome centrally: force updates, set a minimum version, and control which extensions can be installed.
3. Review OAuth app access to Workspace data, and block apps that request broad scopes without a business owner.
4. Set data regions and context-aware access where your edition supports them.

Zoho Workplace

1. Enforce multi-factor authentication for every user and administrator, and remove unused third-party integrations.
2. Confirm your data centre region and export audit logs into your own monitoring.

Apple iWork and iCloud

1. Use Managed Apple Accounts through Apple Business Manager, and turn off iCloud services you don't need.
2. Keep macOS and iOS current, since iWork's security largely rides on the operating system.
3. Use Advanced Data Protection on personal accounts where policy allows, and check its availability for managed accounts with Apple.

LibreOffice

1. Deploy the latest release from the advisories page, and track new advisories as part of your patch cycle.
2. Set macro security to Very high through your deployment tooling, so only macros from trusted file locations run.

ONLYOFFICE

1. Upgrade every self-hosted Docs or DocSpace server today, starting with anything exposed to the internet, and confirm none still runs a version older than the CVE-2021-3199 fix.
2. Enable JWT with a strong, unique secret, put the document server behind a reverse proxy, and limit who can reach it.
3. Watch the project's changelog and releases, and add the server to your external attack-surface checks.

Proton

1. Enforce two-factor authentication across the organization.
2. Teach users which fields are not end-to-end encrypted, such as subject lines, and how to send password-protected mail to outside recipients.

Every suite

1. Write a continuity plan for a day without mail and chat: a second channel for security alerts, a phone tree for incident response, and a way to reset passwords if your identity provider is affected.
2. Warn users after every major outage, because outage-themed phishing follows quickly.
3. Put AI assistants under policy before broad rollout: what data they may see, who may use them, and how prompts and responses are logged.

How Safeguard helps, whichever suite you run

Your suite vendor patches its own products. Safeguard helps you govern everything around them: the vendors, the integrations, the evidence and the AI use.

Third-party and vendor risk. An office suite is one of your most critical vendors, and every add-in and integration is another one. Safeguard's TPRM, Third-Party Risk Manager, lets you onboard vendors, request SBOMs and products from them, score and continuously monitor supplier risk, and apply policies and gates. For background on building the programme, see what third-party risk management is.

SBOM monitoring for add-ins and integrations. Office add-ins, Teams apps, Workspace add-ons and the connectors your teams build are software with dependencies. Safeguard generates and ingests SBOMs in SPDX and CycloneDX, scans the open-source dependencies of the add-ins you build yourself, and enriches vulnerabilities with EPSS and CISA KEV data, so a newly exploited component in an integration you depend on is flagged against the SBOMs you track.

Compliance evidence. Safeguard's compliance and GRC suite covers frameworks such as SOC 2, ISO 27001 and GDPR, with controls, evidence collection, monitoring, tests and questionnaires, so your suite's security settings and vendor reviews become audit evidence rather than screenshots in a folder. Safeguard itself is SOC 2 Type I and Type II audited, with reports available on request.

Asset discovery. You can't patch an Exchange server or an ONLYOFFICE instance you don't know you run. Safeguard's asset discovery finds and inventories your software assets and applies guardrails and enforcement to what it finds.

Policies and gates. You can write organization-wide policies, including from plain-English prompts, and enforce them as gates at build, CI and release. One example: no internal add-in ships with a known-exploited component.

The AI Gateway and Guard guardrails for Copilot and Gemini style use. When teams build their own assistants, agents or integrations on top of large language models, Safeguard's AI Gateway sits on the path to the model and applies guardrails against prompt injection, jailbreaks and personal data leaving in prompts. Guard, Safeguard's runtime protection, enforces guardrails you define inside your running applications, and agent identity tracks which AI agent holds which access. That gives you one policy layer across models, whichever office suite's assistant your staff also use.

Patched open source, fast. For the open-source components inside the add-ins and services you build, Gold Open Source is free and needs no login, and shows affected and fixed versions so you can move to a safe version quickly.

You can see the platform at https://safeguard.sh.

FAQ

Which office suite is the most secure?

None is secure by default in every way. Microsoft 365 has the broadest admin controls but the largest patching burden and the most exploited flaws. Google Workspace has the smallest patching burden but depends on the browser. Proton has the strongest privacy defaults. LibreOffice and self-hosted ONLYOFFICE give the most control and put all patching on you. Pick by where you want the work to sit, then harden.

Why does Microsoft 365 have so many more CVEs than Google Workspace?

Mostly because Microsoft ships much more installed and self-hosted software (desktop Office, Exchange Server, SharePoint Server) and numbers its own bugs at scale, including some cloud bugs since 2024. Google fixes Workspace bugs on its own servers, where nothing needs a CVE for customers to install. CVE counts measure disclosure practice and attack surface, not quality.

How many Microsoft Office vulnerabilities are on the CISA KEV list?

78 entries for Office apps, Outlook, Exchange and SharePoint as of catalog version 2026.11.10, with 5 added in 2025 and 15 so far in 2026. 33 of the 78 are Exchange or SharePoint.

Was the November 2026 Microsoft 365 outage a cyberattack?

Microsoft said there was no indication the outage was the result of malicious or cyber activity. It is separate from the Outlook zero-day CVE-2026-124194. Microsoft had not published a root cause at the time of writing.

Do Copilot and Gemini train on my company's data?

By their own documentation, no. Microsoft says prompts, responses and Microsoft Graph data aren't used to train foundation models, and Google says Workspace data isn't used for model training outside your domain without permission. Zoho and Proton make similar statements for Zia and Lumo. The bigger practical risk is reach: an assistant can surface anything the user is allowed to see, so fix oversharing first.

What should I do this week if I run Microsoft 365?

Patch Office and Outlook on every channel, block outbound TCP 445, turn on NTLM auditing, require MFA everywhere, and review add-ins and app consents.

Sources

Every source linked in this post, in order of first use. Raw data, scripts and the full query list are kept with the post's working files.

• https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-124194
• https://www.cisa.gov/known-exploited-vulnerabilities-catalog
• https://www.aljazeera.com/news/2026/11/13/microsoft-hit-by-global-outages-what-happened-and-why
• https://safeguard.sh/resources/blog/outlook-zero-day-cve-2026-124194-ntlm-leak-and-microsoft-365-outage-november-2026
• https://msrc.microsoft.com/update-guide
• https://nvd.nist.gov/developers/vulnerabilities
• https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json
• https://docs.google.com/spreadsheets/d/1lkNJ0uQwbeC1ZTRrxdtuPLCIl7mlUreoKfSIgajnSyY
• https://www.microsoft.com/msrc/blog/2024/06/toward-greater-transparency-unveiling-cloud-service-cves
• https://www.cve.org/PartnerInformation/ListofPartners
• https://www.microsoft.com/msrc/blog/2025/08/microsoft-bounty-program-year-in-review-17-million-in-rewards
• https://nvd.nist.gov/vuln/detail/CVE-2025-53770
• https://safeguard.sh/resources/blog/sharepoint-toolshell-cve-2025-53770-patch-response
• https://nvd.nist.gov/vuln/detail/CVE-2026-21509
• https://safeguard.sh/resources/blog/outlook-ntlm-hash-leak-zero-day-cve-2023-23397
• https://safeguard.sh/resources/blog/cve-2024-21413-outlook-moniker-link-analysis
• https://cloud.google.com/blog/topics/threat-intelligence/2025-zero-day-review
• https://bugbounty.zohocorp.com
• https://nvd.nist.gov/vuln/detail/CVE-2025-46316
• https://nvd.nist.gov/vuln/detail/CVE-2025-46306
• https://nvd.nist.gov/vuln/detail/CVE-2025-43200
• https://www.libreoffice.org/about-us/security/advisories/
• https://nvd.nist.gov/vuln/detail/CVE-2021-3199
• https://github.com/ONLYOFFICE/DocumentServer/blob/903fe5ab7a275bd69c3c3346af2d21cf87ebeabf/CHANGELOG.md#563
• https://proton.me/support/proton-mail-encryption-explained
• https://www.microsoft.com/en-us/msrc/bounty
• https://www.securityweek.com/microsoft-bug-bounty-program-20-million-paid-to-500-researchers/
• https://bughunters.google.com/about/rules/4849867320328192/cloud-vulnerability-reward-program-rules
• https://blog.google/security/vrp-2025-year-in-review/
• https://www.theregister.com/2025/10/07/google_ai_bug_bounty/
• https://security.apple.com/bounty/
• https://security.apple.com/bounty/categories/
• https://www.onlyoffice.com/blog/2022/02/onlyoffice-is-launching-a-bounty-program-on-hackerone
• https://proton.me/security/bug-bounty
• https://www.theregister.com/2025/01/13/azure_m365_outage/
• https://www.theregister.com/2025/03/03/microsoft_outlook_outage/
• https://status.cloud.google.com/incidents/ow5i3PPK96RduMcb1SsW
• https://www.computerworld.com/article/4020870/microsofts-19-hour-outlook-outage-exposes-fragility-in-cloud-infrastructure.html
• https://www.bleepingcomputer.com/news/microsoft/microsoft-dns-outage-impacts-azure-and-microsoft-365-services/
• https://www.google.com/appsstatus/dashboard/incidents/vagmms1y6yRxwg6RMQZp
• https://proton.me/blog/august-27-outage-incident-report
• https://www.securityweek.com/microsoft-365-outage-disrupts-outlook-exchange-online-and-teams/
• https://learn.microsoft.com/en-us/entra/identity/conditional-access/overview
• https://learn.microsoft.com/en-us/purview/dlp-learn-about-dlp
• https://learn.microsoft.com/en-us/purview/ediscovery
• https://knowledge.workspace.google.com/admin/security/protect-your-business-with-2-step-verification
• https://knowledge.workspace.google.com/admin/security/protect-your-business-with-context-aware-access
• https://knowledge.workspace.google.com/admin/security/about-dlp
• https://knowledge.workspace.google.com/vault/getting-started/vault-overview
• https://knowledge.workspace.google.com/admin/compliance/choose-a-geographic-location-for-your-data
• https://www.zoho.com/know-your-datacenter.html
• https://www.onlyoffice.com/security.aspx
• https://support.microsoft.com/en-us/office/block-or-unblock-automatic-picture-downloads-in-email-messages-15e08854-6808-49b1-9a0a-50b81f2d617a
• https://learn.microsoft.com/en-us/microsoft-365-apps/security/internet-macros-blocked
• https://learn.microsoft.com/en-us/exchange/clients-and-mobile-in-exchange-online/deprecation-of-basic-authentication-exchange-online
• https://techcommunity.microsoft.com/blog/windows-itpro-blog/advancing-windows-security-disabling-ntlm-by-default/4489526
• https://help.libreoffice.org/latest/en-US/text/shared/optionen/macrosecurity_sl.html
• https://helpcenter.onlyoffice.com/docs/userguides/document-editor/tools-and-settings/using-macros
• https://learn.microsoft.com/en-us/microsoft-365-apps/updates/overview-update-channels
• https://learn.microsoft.com/en-us/compliance/regulatory/offering-soc-2
• https://learn.microsoft.com/en-us/compliance/regulatory/offering-iso-27001
• https://learn.microsoft.com/en-us/privacy/eudb/eu-data-boundary-learn
• https://workspace.google.com/learn-more/security/security-whitepaper/page-5/
• https://www.zoho.com/compliance.html
• https://support.apple.com/guide/certifications/apc34d2c0468b
• https://proton.me/blog/soc-2
• https://proton.me/business/trust
• https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy
• https://msrc.microsoft.com/update-guide/vulnerability/CVE-2025-32711
• https://safeguard.sh/resources/blog/echoleak-microsoft-copilot-cve-2025-32711
• https://knowledge.workspace.google.com/admin/gemini/generative-ai-in-google-workspace-privacy-hub
• https://blog.google/security/mitigating-prompt-injection-attacks/
• https://knowledge.workspace.google.com/admin/security/indirect-prompt-injections-and-googles-layered-defense-strategy-for-gemini
• https://help.zoho.com/portal/en/kb/workdrive/zia-ai/articles/zia-privacy-principles
• https://security.apple.com/blog/private-cloud-compute/
• https://helpcenter.onlyoffice.com/desktop/configuration/desktop-AI-agent.aspx
• https://www.libreoffice.org/discover/new-features/
• https://proton.me/support/lumo-privacy
• https://proton.me/blog/proton-scribe-writing-assistant
• https://safeguard.sh/resources/blog/safeguard-tprm-third-party-risk
• https://safeguard.sh/resources/blog/what-is-third-party-risk-management
• https://gold.safeguard.sh
• https://safeguard.sh