evidence
log exports
Screenshots and CSV log exports assembled by hand the week before fieldwork.
a2a receipts
Signed receipts for Agent API, public /invoke, and standard MCP tools/call executions, queryable across the observation period.
audit evidence, not screenshots
SOC 2 for AI agents, backed by signed receipts.
SOC 2 fieldwork is evidence collection. Agent API, public `/invoke`, and standard MCP `tools/call` executions receive signed receipts with caller, skill, input, verified grant, outcome, and timing evidence. Activity traces, artifacts, evaluations, retention, and inventory remain separate controls.
CC6 · CC7 · CC8 · retained · verifiable
0
signed execution entry points
retention floor (days)
0
control areas discussed
the problem
Your agent's audit evidence is a pile of mutable logs.
Most teams walk into SOC 2 fieldwork with agent 'evidence' that is really a stream of application logs plus screenshots collected the week before. Logs are mutable, the control owner also owns the log store, and a shared production key means the record rarely shows which authority permitted an action. An auditor testing CC6, CC7, or CC8 is being asked to trust a record kept by the party being audited — which is exactly what an audit is supposed to avoid.
Mutable logs — no way to prove a line wasn't edited or removed inside the observation window.
Shared production credentials obscure which authority permitted each agent action (CC6).
Log rotation ages evidence out before the Type II window closes (retention gap).
Evidence is hand-assembled per audit cycle instead of produced continuously by the system.
the a2a way
Every agent run, signed into verifiable technical evidence.
A receipt is a first-class primitive on a2a cloud, not a logging add-on. It signs who called, which skill ran, input evidence, verified grants, outcome, and timing. Rich activity, artifact, and evaluation evidence remains separately retained and correlated.
Evidence, not screenshotsAlteration-evident per runAccess mapped to authorityRetention floor enforcedChange and monitoring trailVerifiable outside the vendor
Evidence, not screenshots
SOC 2 fieldwork runs on evidence. Agent API, public `/invoke`, and standard MCP `tools/call` executions receive receipts covering caller, skill, input evidence, verified grant IDs, outcome/result preview, and timing. Rich activity evidence stays separate.
Alteration-evident per run
Changing a signed receipt field breaks Ed25519 verification. Retention and inventory controls are still required to demonstrate that the overall audit population is complete.
Access mapped to authority
The logical-access criteria (CC6) want evidence that authorized actors performed authorized actions. Each receipt signs the caller and grant IDs; the referenced grants provide audience, TTL, file, and tool scope.
Retention floor enforced
Auditors sample a window. The platform enforces a 180-day retention floor on receipts, grants, and audit records, so the evidence for the observation period is still there when fieldwork starts.
Change and monitoring trail
Deploys, grant issuance, revocations, and runs are queryable from the platform's audit surfaces. Run receipts are signed individually; the surrounding audit events support the broader CC7 and CC8 narrative.
Verifiable outside the vendor
The point of an audit is not trusting the party being audited. A receipt's signature is checked independently of the agent runtime — governed agent execution the auditor can verify, not take on faith.
side-by-side
Hand-collected logs vs. signed receipts.
integrity
log exports
Application logs an operator can edit — the control owner also controls the evidence.
a2a receipts
Ed25519-signed receipts. Altering a signed field breaks verification.
access (CC6)
log exports
A shared production API key; who did what is inferred, not recorded.
a2a receipts
Caller and grant IDs are signed; the referenced grant supplies audience, TTL, and tool scope.
retention
log exports
Log rotation quietly ages out records before the audit window closes.
a2a receipts
A 180-day retention floor on receipts, grants, and audit records.
verifiability
log exports
“Trust our logs” — the auditor relies on the vendor's own tooling.
a2a receipts
Signatures verify independently of the runtime that produced them.
dimension
log exports
a2a receipts
evidence
Screenshots and CSV log exports assembled by hand the week before fieldwork.
Signed receipts for Agent API, public /invoke, and standard MCP tools/call executions, queryable across the observation period.
integrity
Application logs an operator can edit — the control owner also controls the evidence.
Ed25519-signed receipts. Altering a signed field breaks verification.
access (CC6)
A shared production API key; who did what is inferred, not recorded.
Caller and grant IDs are signed; the referenced grant supplies audience, TTL, and tool scope.
retention
Log rotation quietly ages out records before the audit window closes.
A 180-day retention floor on receipts, grants, and audit records.
verifiability
“Trust our logs” — the auditor relies on the vendor's own tooling.
Signatures verify independently of the runtime that produced them.
questions
Frequently asked.
How do signed receipts help with SOC 2 compliance for AI agents?
SOC 2 fieldwork is evidence collection against the Trust Services Criteria. Covered execution entry points produce receipts with caller, skill, input hash and result preview, verified grant IDs, outcome/result preview, and timing. Optional activity fields are not populated by default, and retention, inventory, access, and organizational controls remain necessary.
Which SOC 2 Trust Services Criteria do receipts map to?
Receipts can support CC6 logical-access testing because they sign caller and grant IDs, with scope details available from the referenced grant. Run records and surrounding audit events can also support CC7 monitoring and CC8 change evidence. The receipt signature proves the contents of each receipt; it does not by itself prove that no record was removed.
Is a signed receipt enough evidence for a SOC 2 audit?
Receipts are strong technical evidence for the execution fields they cover, and a verifier can check each signature independently. A full SOC 2 report still needs organizational controls, retention, population-completeness evidence, policies, vendor management, and HR controls. Receipts strengthen the run evidence; they do not replace the audit program.
How long are agent audit records retained?
a2a cloud enforces a 180-day retention floor on receipts, scoped grants, and audit records. That covers a typical SOC 2 Type II observation window so evidence for the period is still present when fieldwork begins, rather than aged out by log rotation.
Do I have to build any of this to pass an AI agent audit?
You still need the organizational controls, policies, risk assessment, vendor management, and population-completeness work required by your audit scope. a2a cloud supplies technical primitives — signed receipts, scoped grants, and a retention floor — so your team does not have to assemble the per-run evidence layer from scratch.
keep reading
Related guides.
All guides live in the guides index.
don't trust the agent
Give your auditor verifiable run evidence.
a2a cloud deploys agents with Ed25519-signed receipts on Agent API, public `/invoke`, and standard MCP `tools/call` executions. Scoped grants and retention controls strengthen the technical evidence layer; policies, access reviews, inventory, and risk controls complete the SOC 2 story.