Answer an Inbound Security Questionnaire in Safeguard, Step by Step

To answer an inbound security questionnaire in Safeguard, open Compliance › Questionnaires, choose Import questionnaire and upload the buyer's .xlsx or .csv file (up to 10 MB), map the question and answer columns, then draft answers from your own approved records. Each answer is saved as a draft, reviewed by a person and approved by a compliance reviewer before it goes back to the buyer.

The Questionnaires area is part of the Compliance and GRC suite, which is generally available (GA). AI-drafted answers are available too, so treat every draft as a starting point that a person checks against its sources. This guide walks through the inbound flow, file in and file out, and is plain about what stays with your team.

If you want the preparation side first (building an SBOM for the release in scope, writing VEX, collecting policy evidence), read Prepare for a security questionnaire with Safeguard. This post starts where that one stops: a buyer has sent you a spreadsheet, and it is due back on Friday.

Key takeaways

• Compliance and GRC suite (GA). Questionnaires sits in the suite alongside Frameworks, Controls, Evidence, Policies, Risks, Vendors and the Trust Center.
• AI-drafted answers (GA). "Draft answers with AI" and the per-question "Suggest" action propose answers labelled "AI suggestion, grounded on your records", with a Sources list. Safeguard's offering ties questionnaire auto-answer to the Trust Center.
• Import from the buyer's file. The import takes an .xlsx or .csv of up to 10 MB, and you choose the sheet, header row and columns.
• People decide, assistants do not. Safeguard's approvals docs say compliance decisions are made by a compliance reviewer, that the Auditor role alone is not a reviewer, and that Safeguard's AI assistant and MCP clients can draft and submit but never approve.
• Trust Center (GA) is where answers every buyer asks for can live, so buyers read them before they send a spreadsheet.
• Safeguard does not send anything for you. Sending the file, and standing behind each answer, stay with your team.
• MCP server (GA) and desktop app (GA) appear in the demos at the end. The remediation section covers autonomous remediation (GA, ships switched off), Gold Open Source (GA) and fork and patch (GA).

Why inbound questionnaires take so long

A security questionnaire arrives as a spreadsheet with a few dozen to several hundred rows, each wanting a yes or no, a short explanation and often a pointer to evidence. The work is slow for reasons that have little to do with security.

1. The questions repeat, but the wording does not. The same MFA question shows up in every questionnaire, phrased ten ways.
2. The answers are scattered across policies, scanners, legal and runbooks.
3. The last answer you sent may be stale. If the policy changed since last quarter, copying the old answer sends a wrong one under your company's name.
4. Nobody owns the final yes. Several people edit rows, and the file goes out with nobody having approved the whole thing.
5. The buyer wants their own file back, because their review tool reads that layout.

AI drafting makes the first two problems smaller and the third and fourth worse if you are careless. A model that writes fluent answers from weak sources produces confident text that nobody checks, which is why sources and review states matter more than speed.

Why buyers are required to ask

The questionnaire is not a whim. Several frameworks put an explicit duty on buyers to assess the suppliers they rely on, and those duties are what land in your inbox.

NIST Cybersecurity Framework 2.0. The Govern function has a whole category for cybersecurity supply chain risk management. GV.SC-06 says planning and due diligence are performed to reduce risks before entering into formal supplier relationships. GV.SC-07 says the risks posed by a supplier, its products and services are understood, recorded, prioritized, assessed, responded to and monitored over the course of the relationship. GV.SC-05 says requirements to address supply chain risks are established and integrated into contracts. A questionnaire is the most common way a buyer shows it did GV.SC-06 before signing and keeps doing GV.SC-07 afterwards.

NIST SP 800-161 Rev. 1. NIST's guidance on cybersecurity supply chain risk management (May 2022, updated as of 1 November 2024) tells organizations how to identify, assess and mitigate supply chain risk, including risk from poor development practices at a supplier. Questionnaires are one of the tools buyers use for that assessment; the buyer side is covered in supplier software risk under DORA, NIS2 and M-22-18.

CISA's Secure by Demand guide. CISA's guide for software customers lays out questions that buyers can ask a software manufacturer about its approach to product security. It makes a distinction worth remembering when you answer: enterprise security is how a company protects its own infrastructure, while product security is what the manufacturer does to make the software it ships secure. Many questionnaires mix the two. Good answers keep them apart.

SOC 2 and the Trust Services Criteria. The AICPA's Trust Services Criteria cover security, availability, processing integrity, confidentiality and privacy, and they are what a SOC 2 examination evaluates. Buyers often ask you to state your report status, which is where the most damaging mistake happens: a stated certification that does not exist.

Your answers become part of the buyer's own records, which is why an answer you cannot stand behind is worse than a slow one.

The common questionnaire formats

The Shared Assessments SIG

The Standardized Information Gathering (SIG) questionnaire comes from Shared Assessments, a member organization for third-party risk practitioners. Shared Assessments describes the SIG as a long-running way to bring consistency to third-party risk assessments across many risk domains. It now also offers SIG EV, a digital experience in which the assessor defines preferred responses when the assessment is created and then compares your completed answers against them.

That last point matters. Many buyers do not read every answer. They compare your answer to the one they expected and look at the mismatches. A vague answer that hedges is a mismatch. A clear "No, and here is our compensating control" is easier for them to process than a paragraph that never says yes or no.

The CSA CAIQ

The Consensus Assessment Initiative Questionnaire (CAIQ) comes from the Cloud Security Alliance. CSA's current release is STAR Level 1: Security Questionnaire (CAIQ v4.1), released on 27 January 2026. CSA describes it as a set of Yes/No questions a cloud customer or auditor may ask a cloud provider to check its compliance with the Cloud Controls Matrix (CCM). The CCM itself is a control framework for cloud computing made up of 197 control objectives across 17 domains. A completed CAIQ is also what a provider submits to the STAR Registry for a Level 1 self-assessment.

Because the CAIQ maps to the CCM, its rows carry control IDs and are grouped by domain. Those columns map naturally to the question ID and category columns in Safeguard's import. Safeguard's offering names SIG and CAIQ as the formats its questionnaire auto-answer works from.

The buyer's own spreadsheet

Most questionnaires you receive will be neither. Large buyers build their own because their risk team tiers vendors, their review tool imports a fixed column layout, and their questions reflect their own regulators and past incidents.

Custom spreadsheets break imports in predictable ways: questions on the third sheet, a header row under a logo, section headings mixed in with questions, and a "Vendor Response" column next to a "Buyer Use Only" column you must not touch. The import steps below give you the controls to handle that; the judgement about which column is yours stays with you.

What the Questionnaires area does, by maturity

Before you import: what the drafts will be built from

A drafted answer can only be as good as the records behind it. The Sources list on a suggestion uses five labels, and each maps to a part of the suite:

• Approved policy, from Compliance › Policies.
• Approved answer, from an earlier questionnaire.
• Document, a document you have stored in the suite.
• Control, from a framework you have switched on.
• Evidence, from Compliance › Evidence.

Safeguard's approvals docs explain what "approved" and "accepted" mean for the records underneath. A policy is in force once a reviewer approves it, and an approval covers one version of the content. Evidence starts as submitted and counts once a reviewer accepts it; accepted evidence with an end date becomes expiring inside its reminder window and expired from that date.

So before a big questionnaire, make sure the policies buyers ask about (access control, encryption, incident response, vulnerability management, business continuity) are approved, that the evidence behind your strongest answers is accepted and current, and that the frameworks your buyers care about are switched on. Thin records produce thin drafts, or none.

Step by step: import the buyer's questionnaire

There is no dedicated questionnaire page in Safeguard's public docs today, so the steps below follow the labels in the app. Where a label differs on your screen, the outcome is what matters.

1. Open the area. Go to Compliance › Questionnaires and choose Import questionnaire. (Use New questionnaire instead if you want to type the questions in yourself, for example when they arrived in an email.)
2. Choose the file. Under "Questionnaire file (.xlsx / .csv)", pick the file the buyer sent. Only .xlsx and .csv are accepted, up to 10 MB. Anything else gets "Choose an .xlsx or .csv file.", a large file gets "That file is larger than 10 MB." and a broken one "That file could not be read as a spreadsheet." Save a legacy .xls as .xlsx first, and ask for the spreadsheet version of a PDF questionnaire.
3. Pick the sheet. For an .xlsx workbook, use the Sheet selector to choose the worksheet that holds the questions, not the instructions tab. If the buyer split questions across several sheets, import each sheet as its own questionnaire with the same sender.
4. Set the header row. The Header row selector lists rows with a preview of each, such as "Row 7: ID, Domain, Question, Response, Comments". Pick the row that holds the column names, not a logo or title row above it.
5. Map the columns. Five selectors, each listing columns by letter and header:
   • Question column. The column with the question text.
   • Answer column. Where your answer goes. If the file has no such column, choose Add a new column at the end.
   • Status column. Where the answer's status goes. Again, Add a new column at the end if there is none.
   • Question ID column. Choose None if the file has no IDs. For a CAIQ, this is the control ID column.
   • Category column. None if absent. For a CAIQ, the domain. On a custom file with both "Buyer Notes" and "Vendor Response", your answer belongs in the second.
6. Check the preview. The sheet shows how many questions it found ("214 questions found") and the First questions. If the count includes section headings, or misses rows, change the header row or the question column and look again.
7. Name it and record the sender. Fill in Name (for example "Buyer A vendor review, Q4") and From (the buyer or the contact who sent it).
8. Import. Choose Import questionnaire. You will see "Questionnaire imported. Draft the answers when you are ready." If something fails, the message is "The questionnaire was not saved. Nothing was imported."

Import tips for messy files

Drafting answers: reading a suggestion

Once the questionnaire is open, its header shows a count such as "0 of 214 answered · 0 approved", the sender, and two drafting actions.

• Draft answers with AI drafts the whole questionnaire. While it runs, the button shows progress such as "Drafting 40/214".
• Suggest on a single question drafts just that one. Use it after you fix a source (approve a policy, accept evidence) and want a fresh draft for the rows it affects.

An AI suggestion with sources

A drafted answer appears under the label AI suggestion, grounded on your records, with a confidence figure such as "82% confidence" and a Sources list naming each source's kind (Approved policy, Approved answer, Document, Control or Evidence) and title. The card marks it Not saved yet. Choose Use this suggestion to put it in the answer box, edit it, and then Save draft.

Read the sources before the answer. The sources tell you whether the sentence can be trusted; the fluency of the sentence tells you nothing. A confidence figure is a hint about how well the draft is supported, not a statement that the answer is true for this buyer's scope.

When there is no draft, or only a template

Some questions will come back without an AI suggestion. The app can say why, for example that nothing in your approved policies, approved answers, documents, controls or evidence matches the question. Some suggestions may instead be labelled Low-confidence template match (not an AI answer), with sources under Catalog templates (not your records).

Treat both as signals, not answers. An empty row means a record is missing or not yet approved, or the honest answer is "No" or "Not applicable". A template match describes what a policy could say; it says nothing about what your company does. Use it as a hint about which policy you might be missing.

A worked example

The table below is an illustration for a fictional vendor called Example Co. It is not data from any real customer, and none of it is a statement about Safeguard.

When confidence is low, ask which source is missing rather than rewording the draft until it sounds better.

Writing answers buyers can process

Lead with the direct answer ("Yes", "No", "Partially", "Not applicable"), because buyers comparing against preferred responses look for it first. Add a sentence of scope (which systems, which environments, since when). Point to evidence by name, such as "Information Security Policy, section 4" or "report available under NDA". State limits plainly: if MFA is enforced for staff but optional for end customers, say so, because a buyer who finds the gap later remembers that you hid it.

Keep enterprise security and product security apart, as CISA's guide does: "we encrypt laptops" answers a question about your company, not about your software. Many rows ask the second question in the words of the first.

Review and approval

Each answer carries a status, shown with the labels Draft, Reviewed and Approved. The actions on an answer are Save draft, Mark reviewed, Approve and Return to draft. Answers from before the review flow may show Accepted earlier (not reviewed); treat those as unreviewed and take them through review before you rely on them.

What the approvals docs guarantee

Safeguard's approvals docs set the rules for compliance decisions across the suite:

• A compliance reviewer decides. A reviewer is an administrator (Super Admin, Tenant Admin or Org Admin), a holder of the GRC role or a compliance admin, manager or reviewer role, or anyone granted the compliance review, approve, admin or manage permission.
• An auditor does not decide your programme. The Auditor role alone is not a reviewer.
• The assistant never decides. A change made through Safeguard's AI assistant or an MCP client such as Claude or ChatGPT can draft and submit, but never accept, approve, launch or acknowledge, whoever is signed in behind it.
• Unknown roles are refused. If Safeguard cannot tell who is deciding, it refuses the decision and says why.
• Decisions are recorded with who made them, when and why.

Who should do which step

A split that works for most teams:

• The drafter (a sales engineer or compliance analyst) runs drafting, fills gaps and saves drafts.
• The subject matter owner (the engineer who runs the encryption setup, the owner of incident response) checks each answer in their area against its sources and marks it Reviewed.
• The compliance reviewer approves, checking scope, tone and anything that reads as a commitment or a certification. They answer for the questionnaire if the buyer comes back with a question.

Do not let one person draft, review and approve the same high-risk answer. And treat an approval as covering the words the approver read: if someone changes an approved answer, take it back through review before it goes out. The wider approval model is in who approves what in Safeguard compliance.

Exporting and sending the answers

When the answers are approved, export them from the questionnaire. For an imported file, the export button is labelled (original layout), for example Export XLSX (original layout), because the buyer expects their own spreadsheet back. A questionnaire made in the app has its own export button.

Before you send anything, open the exported file next to the buyer's original and check it the way the buyer will:

1. Every row you answered is there, in the column the buyer reads, and their own columns are untouched.
2. Nothing unapproved is going out. Look at the status column you mapped or added. Anything not approved is either a gap or a decision you have not made yet: approve it, answer "Not applicable" with a reason, or tell the buyer you will follow up.
3. The workbook still works for them. If the buyer's file relies on other tabs, colour coding, drop-down lists or formulas, paste your answer and status columns into their original workbook rather than sending a different-looking file.
4. The sensitive rows read right. Read the answers about certifications, audits, data location and incident notification one more time. These are the rows a buyer's lawyer copies into a contract.
5. You send it. Send the file yourself, from your own email or the buyer's portal. Safeguard does not send it for you.

Building an answer library you can reuse

The point of approving an answer is that you should not have to write it again. "Approved answer" is one of the source labels a suggestion can cite, so a well-kept set of approved answers is what makes the next questionnaire faster.

Habits that keep the library useful:

• Approve answers that are right. An answer left in Reviewed is a decision nobody finished.
• Re-check after a policy changes. A policy approval covers one version, per the approvals docs. When a policy gets a new version, the answers that cite it may be out of date. Run Suggest on those questions in the next questionnaire and compare.
• Write answers that travel. An answer that says "as discussed with your team on the call" will be wrong in a questionnaire from someone else. Keep buyer-specific detail out of answers you want to reuse.
• Date anything that changes. Audit status, test dates and vulnerability counts age fastest. Put the date in the answer.

Over a few quarters, the work shifts from writing to checking: a person still reads every answer, but reads a sourced draft instead of a blank cell.

When to point the buyer to the Trust Center instead

Some questions are the same for every buyer: audit status, subprocessors, vulnerability disclosure contact, last penetration test. Answering them in every spreadsheet is how inconsistent answers creep in.

The Trust Center (GA) is a public page that carries those standing answers. Per Safeguard's docs, it can show framework statuses, control areas read live from continuous monitoring, policies, subprocessors, an FAQ, dated updates and an audit outcome recorded by your auditor. Its Documents section has a Questionnaire category, so a completed CAIQ or your standard questionnaire can sit there as a document, either Public or Restricted. A restricted document is released only through an approved access request, which a compliance reviewer approves unless an auto-approval rule you wrote decides it.

Point the buyer to the Trust Center when:

• the question is one your page already answers, such as subprocessors or how to report a vulnerability;
• the buyer wants a document (a report, a pen test summary, a completed standard questionnaire) rather than a custom answer;
• you want the buyer to see a live status, not a snapshot that goes stale the day after you send it.

Keep answering in the spreadsheet when the buyer's review tool needs their own layout, when the question is specific to their deployment, or when their contract requires answers in their format.

The docs also set a rule worth copying into your answers: a framework that needs an independent auditor or certification body cannot be shown as compliant on a self-assessment. If your Trust Center would not show it as compliant, your questionnaire should not say it either. The setup steps are in Setting up a Safeguard Trust Center, step by step.

What must never be sent without a person

The OWASP Top 10 for LLM Applications 2025 lists misinformation as LLM09. OWASP describes two causes that apply directly here: hallucination, where a model produces content that sounds right but is unfounded, and overreliance, where people trust model output without checking it. A questionnaire is a written statement to a buyer under your company's name. Overreliance there is not a technical problem; it is a commercial and legal one.

Never send any of these without a named person approving them:

1. Unreviewed AI drafts. A draft is a proposal, not your company's answer.
2. Template matches. They describe a catalogue policy, not your company.
3. Claims about certifications, attestations or audits you do not hold. "SOC 2 compliant" when the audit is in progress, "ISO 27001 certified" when you have a gap assessment, "FedRAMP authorized" when you have an architecture designed toward it. Write the true state on the day, and date it.
4. Commitments about the future. A roadmap date becomes a contract term once the buyer's lawyer reads it.
5. Numbers without scope or date. "Zero critical vulnerabilities" is true for one product on one day. Say which product, which scan and which date, or do not give the number.
6. Anything with a confidence figure you did not check. Confidence describes the draft's support, not the truth of the answer.

A useful test before approving: if the buyer asked you to show the evidence for this answer tomorrow, could you? If the sources cannot support the sentence, rewrite the sentence.

Decision table: where each answer should come from

Where an assistant fits

If your team works in Claude, connect Safeguard through the connector directory, where it is listed as a community connector: open the directory entry, connect, sign in to Safeguard, check the app name and host, and choose Authorize access. On a Team or Enterprise plan, an Owner adds it first.

The connector reads data already in your account, such as products, projects, vulnerabilities and, where they are returned, findings. That helps with a row like "How many open critical vulnerabilities affect the product in scope?" Which tools appear depends on your plan and your administrator's settings, and the approvals docs are clear that an assistant cannot approve anything. Paste what you verify into the answer box, save it as a draft and let the review flow do its job. To count correctly, see how Safeguard data is structured.

What Safeguard does not do

• It does not send the questionnaire. Sending the file, uploading it to the buyer's portal and following up stay with your team.
• It does not make AI drafts correct. A draft comes with its sources, but a person still has to check that the sources support the sentence and that the scope is right.
• It does not fill gaps in your programme. If a policy, test or report does not exist, no draft can make it exist. Thin records produce empty or weak rows; the fix is better records, not a cleverer prompt.
• It does not read every file format. Import accepts .xlsx and .csv up to 10 MB. Legacy .xls, PDF, Word documents and buyer web portals are outside the import.
• It does not let an assistant or an auditor approve. Per the approvals docs, a compliance reviewer decides; assistants and MCP clients can prepare work but never approve, and the Auditor role alone is not a reviewer.
• It does not certify you. Nothing in the suite is an audit opinion or a certificate. A readiness score is Safeguard's assessment of what it can see. An accredited auditor or certification body issues any report or certificate, and your answers must say so honestly.

Common mistakes

1. Importing the instructions tab. If the preview shows "Please complete all fields", pick the sheet with the questions.
2. Mapping the answer to the buyer's column. Their "Assessor Comments" column is not yours.
3. Approving in bulk without reading the sources. Approval is per answer for a reason. Reading the source list takes seconds and catches the answer that cites a superseded policy.
4. Rewording a low-confidence draft until it sounds right. If the confidence is low, a source is missing. Add or approve the source, then run Suggest again.
5. Treating a template match as your policy. The label says "not an AI answer" and "not your records". Believe it.
6. Sending answers nobody approved. Check the status column before the file leaves.
7. Leaving good answers in Reviewed. If a reviewed answer is right, approve it so it can serve as an approved answer next time.
8. Writing buyer-specific detail into reusable answers. It ends up in someone else's questionnaire.
9. Stating a certification you do not hold. Write the true status on the day. "Audit in progress" is a respectable answer. "Certified" when you are not is the answer that ends a deal later, or worse.

FAQ

Can Safeguard fill in a security questionnaire automatically?

It can draft answers for every question at once with "Draft answers with AI", each shown with its sources. It cannot decide for you or send the file. A person saves and reviews each answer, and Safeguard's approvals docs say compliance decisions are made by a compliance reviewer, never by the AI assistant.

Which file formats can I import into Safeguard Questionnaires?

.xlsx and .csv, up to 10 MB per file. You pick the sheet for a workbook, the header row, and the question, answer, status, ID and category columns. Save a legacy .xls as .xlsx first, and ask the buyer for a spreadsheet if they sent a PDF.

Does Safeguard support SIG and CAIQ questionnaires?

Safeguard's offering names SIG and CAIQ as the formats its questionnaire auto-answer works from, and you import them like any other .xlsx or .csv. A CAIQ's control ID and domain columns map naturally to the question ID and category columns. Safeguard does not submit anything to the STAR Registry or to the buyer for you.

Where do the AI drafted answers come from?

The Sources list on each suggestion shows what it drew on: approved policies, approved answers from earlier questionnaires, documents, controls and evidence. Check those sources before you save the draft.

Who can approve questionnaire answers in Safeguard?

Per Safeguard's approvals docs, a compliance reviewer: an administrator, a holder of the GRC role or a compliance admin, manager or reviewer role, or someone with the compliance review, approve, admin or manage permission. The Auditor role alone is not a reviewer, and AI assistants and MCP clients never approve.

How do I make sure nothing unapproved reaches the buyer?

Map or add a status column at import, then check it in the exported file before you send. Anything not approved is either a gap to close or a decision to make, and you send the file yourself.

How do approved answers get reused?

"Approved answer" is one of the source labels a suggestion can cite, so approved answers from earlier questionnaires can support drafts for the next one. Keep them current: re-check them when the policy behind them changes.

Should I send a questionnaire or point the buyer to a Trust Center?

Both, usually. Put standing answers and documents on a Trust Center so buyers can read them first, and answer the buyer's own spreadsheet for anything specific to their deployment or required in their format.

Answering the remediation rows honestly

Many questionnaire rows ask how quickly you fix vulnerable dependencies, and the honest answer depends on what your pipeline does when a fix exists and when it does not. The block below describes Safeguard's remediation options and their maturity, so you can answer those rows from what your team actually runs.

Fix it without waiting for upstream

Whatever the vulnerability, most of the risk sits in two gaps. The first is the time between a fix being published and that fix actually running in your production builds. The second, and worse, is the time before any fix exists at all, while a maintainer works out a patch and a release schedule. Waiting through both gaps is a habit, not a requirement. Safeguard gives teams three ways to close them.

1. Autonomous remediation: pull requests, not tickets

When a fixed version exists, Safeguard's Griffin models open a tested, policy gated pull request that moves the affected dependency to the safe version. You can run it on one repository or in bulk across every project that carries the vulnerable version. Remediation categories include dependencies, licenses, security posture and findings, and each one is set separately. Reachability analysis ranks the vulnerable code paths your application actually calls first, so the first pull requests you merge are the ones that matter most. Autonomous remediation ships switched off, with every category set to Manual, so your team decides how much runs on its own and what still needs a human approval.

2. Gold Open Source: the safe version in seconds, free

gold.safeguard.sh is free and needs no login. Search any CVE or package to see affected and fixed versions, CVSS, EPSS, CISA KEV status and a recommended safe version, or use the "Am I affected?" checker. The same data is available through a free read API, README status badges and RSS threat feeds.

3. Gold Registry: start from components that are already fixed

Gold Registry provides hardened, drop in replacements for upstream packages and container images. Every Gold artifact is held continuously to zero exploitable critical or high CVEs and a completed license check, with signed build provenance (signed with Safeguard's key and verifiable with cosign, SLSA Build Level 3 as the design target) and a public evidence page you can hand to an auditor. An artifact loses its Gold badge if it regresses. Teams can point their own package directory at the registry so hardened equivalents resolve without changing how developers install anything.

When upstream has no fix yet, or will not ship one, Safeguard first works to get the fix accepted upstream. If the issue is severe or upstream will not move, Safeguard builds a fork, fixes the vulnerability there and publishes a hardened Gold build, so you are not stuck on a maintainer's release schedule. You can also request a hardened, compatibility tested build of a specific package. Ask which packages it covers before you plan around it.

One honest limit: when the vulnerable code lives in a vendor appliance, a hosted service or an operating system runtime you do not build yourself, the fix still has to come from that vendor or distribution. The steps above apply to the open source dependencies and container images you build and ship, which is where most teams carry the largest and least tracked share of their exposure.

See how Safeguard would help

The first demo connects Claude to the Safeguard MCP server with OAuth and shows how you choose which tools Claude can use. While you answer a questionnaire, the same connection lets an assistant look up the numbers a row asks for, such as open critical vulnerabilities in the product in scope. It can read and summarise, but approving answers stays with a compliance reviewer in the app.

Click through it at your own pace, or open it full screen.

The second demo is a real session in the desktop app. The CLI session shows in the conversation list, you ask for work on your computer, pick the folder, then approve or deny the commands and file edits it proposes, and it ends with an upgrade done and tests passing. When a questionnaire row asks how fast you fix a vulnerable dependency, this is the kind of reviewed fix flow you can describe, because a person approves each change.

Click through it at your own pace, or open it full screen.

To try the flow yourself, get Safeguard from safeguard.sh/download and read the Safeguard docs for the compliance suite and the Trust Center.

Quick check

Sources

• Safeguard docs: Compliance
• Safeguard docs: Trust Center setup
• Safeguard docs: Approvals
• Safeguard docs: Compliance frameworks
• Safeguard docs: Connect Claude to Safeguard
• Safeguard blog: Prepare for a security questionnaire with Safeguard
• Cloud Security Alliance: STAR Level 1 Security Questionnaire (CAIQ v4.1)
• Cloud Security Alliance: Cloud Controls Matrix
• Shared Assessments: SIG and SIG EV
• NIST: The Cybersecurity Framework (CSF) 2.0, CSWP 29
• NIST SP 800-161 Rev. 1: Cybersecurity Supply Chain Risk Management Practices
• NIST SP 800-218: Secure Software Development Framework
• CISA: Secure by Demand Guide
• AICPA: Trust Services Criteria
• OWASP Top 10 for LLM Applications 2025: LLM09 Misinformation

Last reviewed: 24 October 2026. Product capabilities change; check docs.safeguard.sh for the current state.