Every organisation
AI security questionnaire pack: evidence for customer AI vendor risk assessments
An AI security questionnaire is the set of questions a customer sends before buying, asking how your AI tools handle their data, what is logged, which actions AI can take and how incidents are handled. The Cloud Security Alliance's AI-CAIQ is a standard version built on its AI Controls Matrix; most customers add questions of their own.
Not sure this rule covers you? See who each rule covers
The rule
Your customers' security questionnaires: contractual, not law
- Who it covers
- Any company that sells to other businesses, and most of all to banks, insurers and other regulated firms.
- What it asks
- Answer questions about AI: what is logged and for how long, how personal data is protected, how incidents are handled, and which tools an AI system can use.
- When
- At each customer review.
The questions that come up
Security teams at banks, insurers and larger companies now ask about AI in every review. The questions cluster into six groups:
- Inventory: which AI systems and models do you use, including the AI apps your staff run on their own laptops?
- Data: what customer data can an AI tool read, is it sent to a model provider, is it used for training, and how is personal data protected?
- Actions: what can an AI system do, not just read? Can it send email, change records, move money? Who approved that list?
- Logging: is every AI action recorded, with what detail, for how long, and can you produce the record for a specific customer or incident?
- Access and secrets: how are the credentials AI tools use stored, scoped and revoked?
- Incidents: how would you detect an AI system doing something it should not, and who is told?
What to attach as evidence
Answers that point to evidence close faster than answers that describe intent. For each group, the pack contains:
- Inventory: the list of AI apps and API keys that use your workspace, with who signed each one in and when it was last seen.
- Data: the personal-data types redacted from tool results before an AI app sees them, with counts over the review period.
- Actions: the tools each AI app can use, connection by connection, and the tools switched off.
- Logging: a sample of the activity record showing the AI app, user, tool, arguments, redacted result and policy outcome for a call, plus the CSV export.
- Access and secrets: where the credentials for connected apps are held, and that tool calls carrying secrets are blocked.
- Incidents: the blocked and warned calls over the period, and the policies that fired.
Answering "what can your AI actually do?"
This is the question most vendors answer badly, because the honest answer in most companies is "whatever the MCP servers on each laptop expose". A single governed connection changes the answer to a list: these apps, these tools, switched on for these AI apps, with these policies on every call. That list, and the record that shows it being applied, is what the reviewer is looking for.
ContextGate gives you the list and the record for the tool calls that go through it. The answers themselves, and the policies behind them, are yours to write. It does not cover an AI app's built-in tools or the chat itself, and it is not a certification.
The report
AI security Review Pack
CSA AI-CAIQ and customers' own questionnaires: contractual, not statutory
Answers to the AI questions in your customers' security reviews: what is logged, how personal data is protected, and which tools an AI system can use.
What it covers
- What is recorded for each tool call: the AI app, user, tool, arguments and policy outcome.
- Personal data redaction, secret blocking and switched-off tools, as controls in force.
- Which tools each AI app can use, connection by connection.
Who needs it
Any company that sells to other businesses, and most of all to banks, insurers and other regulated firms.
What you still do
Write the answers and the policies behind them.
Questions people ask
What is an AI security questionnaire?
A set of questions a customer's security or procurement team sends to a supplier about the AI systems the supplier uses or sells: data handling, logging, permitted actions, access to credentials, incident handling and governance. It is usually part of a wider vendor risk assessment.
What is the CSA AI-CAIQ?
The Cloud Security Alliance's Consensus Assessment Initiative Questionnaire for AI, built on its AI Controls Matrix. It gives a standard list of yes/no questions mapped to controls, so that one completed questionnaire can answer many customers.
What evidence should an AI vendor risk assessment include?
An inventory of AI systems, the data each can reach, the actions each can take, a sample of the activity record with the detail logged for each call, the retention period, where credentials are held, and the controls that block or redact, with evidence of them operating.
How do you prove which tools an AI system can use?
With a per-tool list for each AI app and connection, and a record of the calls actually made. A gateway in front of the tools can produce both, because every call passes through it and the tools switched off are never offered to the AI app.
Does passing a security questionnaire make us compliant?
No. A questionnaire is contractual, between you and one customer. It helps you become compliant with their requirements and gives you a reusable pack for the next review, but it is not a certification or a legal finding.
A summary, not legal advice: check your obligations with your compliance team or adviser.
Ready to see what your AI agents do?
Create a free workspace and add ContextGate to your AI apps in minutes. No card required.
Get started free
We'll email you a sign-in link to set up your workspace.
Rather talk it through first? Book a 30-minute call