Evidence checklist

Security questionnaire evidence checklist

Map common customer security questions to accepted evidence, stronger proof, weak evidence signals, owners, and review cadence.

Download CSV Build answer library Download evidence request email

Best next stepUse before sending answers

Treat every customer-facing claim as unsupported until it has an owner, source, last review date, and customer-safe evidence format.

Minimum evidence pack by risk tier

Use the smallest evidence pack that matches the customer, data access, and vendor risk. This keeps questionnaires shorter without weakening proof.

Low-risk review

Security policy, access control summary, encryption statement, incident response summary, subprocessor list.

Medium-risk review

Low-risk pack plus SOC 2/ISO evidence, access review record, pentest summary, DPA, retention/deletion process.

High-risk or AI review

Medium-risk pack plus AI data use statement, model provider review, audit log sample, OAuth scope map, token revocation runbook, tenant isolation evidence.

Minimum evidence pack before you answer an AI vendor or MCP review

Use this when the buyer is really asking for retrievable proof, not a long narrative. Each path should end with evidence a reviewer can open, inspect, and reuse.

AI vendor reviewAI data use statement, model provider or subprocessor list, training and retention position, human review owner, and customer-safe evidence link.

MCP / agent reviewTool permission map, minimum OAuth scopes, token revocation runbook, approval path, audit log sample, and tenant-isolation note.

Customer security reviewSecurity policy summary, access review proof, encryption scope, incident response summary, and the approved answer-library row that cites each source.

AI request boundary evidence

Use this table when a customer asks how AI, MCP, or agent actions are controlled. The answer should prove the boundary around each request, not just list the tools.

What proof customers acceptUse when a self-written AI policy is not enough

A customer-safe AI evidence package should include model and provider list, touched data classes, training and retention stance, subprocessors, human review rule, prompt-injection handling, per-request audit sample, and disable or rollback path.

Identity contextWhich user, tenant, role, service account, or agent identity made the request and which customer data boundary applied.

Policy decisionThe rule, approval workflow, or policy decision point that allowed, denied, scoped, or escalated the AI or agent action.

Prompt injection handlingHow untrusted instructions, retrieved content, tool descriptions, and customer inputs are filtered, isolated, or reviewed.

Per-request audit trailTimestamp, actor, tool or model, requested action, data touched, policy result, reviewer if any, and retention period.

Provider and retention boundaryModel provider, subprocessors, training position, prompt/output retention, deletion path, and customer opt-out settings.

Disable or rollback pathHow a risky model, MCP server, tool permission, token, or agent workflow can be revoked and how that action is evidenced.

Evidence map

Use this as a source checklist for answer libraries, customer questionnaires, DDQs, SIGs, CAIQs, and AI vendor reviews.

Area	Customer question	Accepted evidence	Stronger evidence	Weak evidence signal	Owner
Security programAnnual	Do you maintain a formal information security program?	Security policy, control owner list, risk register, annual review record	SOC 2 report, ISO 27001 certificate, board or leadership review evidence	A short marketing statement without policy owner, date, or scope	Security / Operations
Access controlQuarterly	How do you control and review employee access?	Access control policy, IdP groups, access review export, offboarding checklist	Quarterly access review evidence, privileged access logs, SSO/MFA enforcement evidence	Saying access is role-based without a review record	IT / Security
EncryptionSemiannual	Is customer data encrypted in transit and at rest?	Encryption policy, TLS documentation, cloud encryption configuration, architecture note	SOC 2 section, key management documentation, exception register	Generic cloud provider encryption claim with no product scope	Engineering
Incident responseAnnual	Do you have an incident response and notification process?	Incident response policy, escalation contacts, tabletop notes, notification assessment process	Recent tabletop result, post-incident review template, legal-approved customer notification language	Promise of fast notification without legal-approved trigger or timeline	Security / Legal
Vulnerability managementQuarterly	Do you scan, test, and remediate vulnerabilities?	Vulnerability management policy, scan cadence, remediation SLA, pentest summary	Customer-safe pentest attestation, remediation tracker, severity-based SLA history	Raw pentest report shared without remediation status or scope explanation	Security / Engineering
Privacy and data processingSemiannual	What personal data do you process and retain?	Privacy policy, data inventory, retention policy, DPA, subprocessor list	DPIA / privacy risk assessment, deletion workflow evidence, data flow diagram	Privacy policy link that does not cover product processing or retention	Privacy / Legal
Third-party riskAnnual	How do you assess vendors and subprocessors?	Vendor risk assessment, DPA, SOC 2 review note, subprocessor approval record	Risk-tiered supplier review, renewal review evidence, critical vendor exit plan	A vendor list with no data access or criticality classification	Security / Procurement
AI vendor reviewQuarterly	How do AI vendors process prompts, outputs, and customer data?	Model provider terms, DPA, retention statement, no-training commitment where applicable	AI vendor risk assessment, deletion process, subprocessor mapping, human review notes	Saying AI is secure without provider-specific evidence	Product / Security / Legal
MCP and agent toolsQuarterly	How do you govern MCP servers, tool calls, OAuth scopes, and token revocation?	MCP security checklist, OAuth scope map, approval workflow, token revocation procedure	Audit log sample, tenant isolation test, tool permission review, prompt injection control evidence	Listing MCP tools without explaining permissions, logging, or revocation	Security / Platform

Strong evidence, acceptable evidence, and weak evidence

AI Overviews and customer reviewers both reward specificity. Use this table to separate evidence you can safely attach from claims that need review before reuse.

Area	Acceptable evidence	Stronger evidence	Weak evidence	Do not claim unless verified
Access control	SSO/MFA configuration, access review export, offboarding checklist	Quarterly access review with reviewer, date, scope, and exceptions	A statement that access is role-based without review evidence	Do not claim quarterly review if the last review cannot be produced.
Encryption	Architecture note, KMS/key management policy, TLS configuration summary	SOC 2 control reference plus product-specific encryption scope	Generic cloud-provider encryption language with no product boundary	Do not imply all exports, logs, backups, or customer-managed files are covered unless verified.
Incident response	IR policy, escalation owner, notification assessment process	Tabletop record, post-incident template, legal-approved notification language	A promise to notify immediately without trigger criteria	Do not promise a notification timeline that conflicts with contracts or legal review.
Subprocessors	Public subprocessor list, DPA, data flow summary	Criticality tiering, access scope, region, and change-notification process	A vendor list with no data-access explanation	Do not say no third parties access data if infrastructure, support, AI, or analytics vendors can process it.
AI data use	Model provider terms, no-training statement, retention note, human review owner	AI vendor assessment with prompt/output retention, deletion, subprocessor, and audit-log evidence	Saying AI is secure without provider-specific data handling proof	Do not answer AI training questions from memory; verify provider terms and product settings.

15 evidence-led questions (short questionnaire)

Community discussions often argue that shorter questionnaires work better when every question forces an evidence artifact, not a yes/no.

Question	Evidence to request
Share a high-level architecture + data flow diagram.	Architecture/data flow diagram + data locations
List subprocessors that can touch customer data and where the canonical list lives.	Public subprocessor list + internal register
Where is data stored/processed (regions) and how do you handle data residency requests?	Data residency statement + subprocessor regions
How do you enforce SSO/MFA and manage joiner/mover/leaver access?	IdP config + offboarding checklist
How often do you run access reviews and what proves the last review happened?	Access review export + owner + date
How do you isolate tenants and prevent cross-tenant access?	Tenant isolation design note + test evidence
What audit logs exist, who can access them, and how long are they retained?	Audit logging policy + log sample + retention
What is your vulnerability management process and remediation SLA?	VM policy + SLA + tracker evidence
Do you have a recent penetration test summary and a remediation status note?	Customer-safe pentest summary + remediation status
What is your incident response process and notification timeline?	IR plan + notification decision template
How do you back up data and test restores (RPO/RTO evidence)?	Backup policy + restore test record
What personal data do you process and how long do you retain it?	Data inventory + retention schedule
How do you delete customer data and prove deletion on request?	Deletion runbook + sample completion record
If you use AI, do providers train on customer data and what are retention/opt-out controls?	AI vendor terms + DPA + retention controls
If you use MCP/agents, what tool permissions, OAuth scopes, and token revocation controls exist?	OAuth scope map + approval + revocation evidence

Score vendor risk Review MCP gateway evidence Add privacy review fields Evidence pack is ready, score my process

Evidence rules

Good evidence reduces back-and-forth with customers and makes automation safer later.

Use customer-safe proof

Share summaries, attestations, policy excerpts, trust-center links, or controlled access instead of raw internal artifacts.

Keep dates visible

Track last reviewed and next review dates so repeated answers do not drift away from current controls.

Separate claims from caveats

Use exception notes for roadmap items, partial controls, compensating controls, and customer-specific promises.

Tie AI to vendor review

For AI providers and MCP tools, document data use, retention, scopes, logs, tenant isolation, and revocation.

Evidence checklist FAQ

Short answers for teams deciding what to attach, what to hold back, and what needs owner review.

What evidence should I attach to security questionnaire answers?

Attach customer-safe proof that matches the claim: policies, SOC 2 sections, ISO certificates, access review records, architecture notes, DPA/subprocessor lists, incident response summaries, audit log samples, and AI vendor terms where relevant.

Is SOC 2 enough for a security questionnaire?

SOC 2 helps, but it usually does not answer environment-specific questions about data residency, subprocessors, deletion, access reviews, AI providers, tenant isolation, or customer-specific evidence.

What if we do not have formal evidence yet?

Answer with the control that exists today, mark the claim level as partial or roadmap, attach a safe substitute such as a policy excerpt or owner note, and avoid claiming a certification, report, or test that does not exist.

What is weak evidence in a customer security review?

Weak evidence is a broad marketing statement, outdated policy, screenshot with no owner/date, generic cloud-provider claim, or yes/no answer that cannot be traced to a source.

Where should AI approval evidence live?

Keep one retrievable review record per approved AI tool or workflow: decision, owner, risk tier, approval conditions, review date, and links to the supporting vendor, privacy, and security evidence.