← Standards register Assessment

AIUC-1

Agent certification · 51 controls in six domains

The two controls this standard cares most about are stopping unauthorized agent actions and unsafe tool calls. Provenant handles both, and neither can be turned off by a configuration mistake.

Ref	Requirement	How	Mechanism
A1	Establish input data policy	Enforced	You choose where every value comes from
A2	Establish output data policy	Enforced	You choose what the agent sees
A3	Limit AI agent data access	Enforced	You choose what the agent sees, field by field
A4	Protect IP and trade secrets	Enforced	The same control, applied to anything you mark proprietary
A5	Prevent cross-customer data exposure	Enforced	One customer's session can't reach another's
A6	Prevent PII leakage	Enforced	You choose what the agent sees, field by field
A7	Prevent IP violations	Contained	Your agent can only send things where you allowed
A8	Prevent leakage of credentials and secrets	Enforced	Your agent uses credentials without ever seeing them
B1	Third-party testing of adversarial robustness	Evidenced	Your testers get the full list of what to test
B2	Detect adversarial input		N/A
B3	Manage public release of technical details		N/A
B4	Prevent AI endpoint scraping		N/A
B5	Implement real-time input filtering		N/A
B6	Prevent unauthorized AI agent actions	Enforced	Your agent can't do what you didn't allow
B7	Enforce user access privileges to AI systems	Enforced	Who the agent acts for is settled before it starts
B8	Protect AI system deployment environment		N/A
B9	Limit output over-exposure	Enforced	Your agent gets what you chose and nothing else
B10	Promote secure patterns in generated code	Contained	Generated code can only reach what you allowed
C1	Define AI risk taxonomy	Evidenced	Your risk categories come from the data you actually handle
C2	Conduct pre-deployment testing	Enforced	You can see everything your agent could do before you ship it
C3	Prevent harmful outputs	Contained	A bad answer can only reach what you allowed
C4	Prevent out-of-scope outputs	Enforced	Your agent can't step outside what you allowed
C5	Prevent agent-specific high-risk outputs	Enforced	The steps that matter stop for a person
C6	Prevent output vulnerabilities	Contained	Limited to what you allowed
C7	Flag high-risk outputs for human review	Enforced	You choose what the approver sees before they decide
C8	Monitor AI risk categories		N/A
C9	Enable real-time feedback and intervention	Enforced	A person can step in at the moment it matters
C10	Third-party testing for harmful outputs	Evidenced	Your testers get the full list of what to test
C11	Third-party testing for out-of-scope outputs	Evidenced	Your testers get the full list of what to test
C12	Third-party testing for customer-defined risk	Evidenced	Your testers get the full list of what to test
D1	Prevent hallucinated outputs	Contained	A wrong answer can only reach what you allowed
D2	Third-party testing for hallucinations	Evidenced	Your testers get the full list of what to test
D3	Restrict unsafe tool calls	Enforced	What your agent can do next depends on what it has done and who approved it
D4	Third-party testing of tool calls	Evidenced	The tests come straight out of what you allowed
E1	AI failure plan for security breaches	Evidenced	Your plan gets both halves — what could have happened, and what did
E2	AI failure plan for harmful outputs	Evidenced	The same two halves
E3	AI failure plan for hallucinations	Evidenced	The same two halves
E4	Assign accountability	Evidenced	You know who approved what, and what they saw
E5	Document data storage security	Evidenced	You choose what is sent and what is kept
E6	Conduct vendor due diligence	Evidenced	Your vendor list writes itself
E7	Review internal processes	Evidenced	Every change to what your agent can do is tracked and attributable
E8	Monitor third-party access	Enforced	A signed record of every access
E9	Establish AI acceptable use policy	Evidenced	Your policy and your enforcement are the same document
E10	Record processing locations	Enforced	Recorded for every system your agent touches
E11	Document regulatory compliance	Evidenced	Written for you
E12	Implement quality management system	Evidenced	Every change is tracked and attributable
E13	Log AI system activity	Enforced	Every decision recorded as it happens, and it can't be altered later
E14	Implement AI disclosure mechanisms		N/A
E15	Document system transparency policy	Enforced	The document is written for you, and it is the policy
F1	Prevent AI cyber misuse	Contained	Limited to what you allowed
F2	Prevent catastrophic misuse	Contained	Limited to what you allowed