AgentProof Living Intelligence
How AgentProof stays current as the AI-agent landscape moves
Guidance, security, platform and governance signals change constantly. AgentProof tracks them — but nothing reaches your report until a human reviews and approves it.
What Radar is, and what keeps your records current
Radar is a deliberate review cadence plus a versioned methodology change log that keeps AgentProof current as AI-agent guidance, security expectations, platform updates, and governance signals move. It keeps a compliance-readiness record from going stale the day after it prints.
Nothing applies automatically. Every signal is reviewed by a named human and pinned to a release before it can change a question, scoring, or methodology. It is a review process, not a live scan; there is no autonomous or continuous scanning. AgentProof does not call any live AI provider when it scores your agent, and it never rewrites your scores or your frozen documentation packs on its own.
Why the signals change
Four kinds of signal move the goalposts for what a ready AI agent looks like. AgentProof watches each of them factually — what changed, and why it matters to your report.
What changes
Vendor guidance for building agents — the agent platforms and workflow tools your teams use — is updated regularly, naming new concepts and recommended controls.
Why it matters
Agents built on a platform tend to follow that platform's current guidance, so the expectations a buyer is held to move with it.
What changes
Security references evolve — the OWASP LLM Top 10, OWASP Agentic work, MITRE ATLAS, and observed prompt- and tool-injection patterns each add new findings over time.
Why it matters
Tool and API surfaces are where most agentic impact happens, so testing expectations broaden as new patterns are documented.
What changes
Model providers and platforms change guardrails, permission models, availability, and capabilities; AgentProof tracks the named concepts, not marketing.
Why it matters
A change to a permission model or a guardrail can change what a 'ready' agent looks like on that platform.
What changes
Governance references phase in over time — the EU AI Act phases in legal obligations, while the NIST AI RMF and ISO/IEC 42001 add named, voluntary expectations.
Why it matters
Buyers in governed contexts increasingly expect documented oversight and transparency mapped to these references.
The human-approval pipeline
Here is exactly how a signal travels from an external source to a customer-visible change, seven human-reviewed steps from source check to reassessment trigger.
1. Source check
An approved source is checked. A validation read never changes anything and never leaves the server.
2. Signal extraction
The fetch is reduced to one or more candidate signals — short, structured records describing what changed.
3. Status assignment
Each signal lands in 'Watching', 'Under review', 'Approved', or 'Affects reports'.
4. Proposal
Approved signals become proposals — concrete suggestions for methodology / scoring / question changes.
5. Human approval gate
No proposal modifies the methodology pack until a human admin approves it. Reviewers operate on an internal, founder-only console.
6. Publication
Approved proposals are published as a new intelligence pack version. Old reports continue to reference their original pack version.
7. Reassessment trigger
When a published change affects an existing report's control, the customer sees a 'Reassessment recommended' notice with a link to review and re-run when they choose.
No silent drift
Every recommendation cites the intelligence-pack version it came from. Saved reports stay pinned to the version they were scored under; when the methodology changes, affected reports show a ‘Reassessment recommended’ banner — so you always know what changed and when.
‘Reassessment recommended’ is a customer-facing notice, never an automatic re-score. You stay in control of when, and whether, to re-run a report against a newer methodology version.
Honest scope — what this is, and is not
Review-ready
Every recommendation cites the intelligence-pack version it came from. Signals are tied to named, approved references — not a live web crawl.
Human-reviewed
A named human reviews and approves every methodology, scoring, or question change before it goes live. No signal changes your score automatically.
Honest scope
Today the radar is internal human review, not autonomous continuous scanning. Curated intelligence - human-reviewed today, with scheduled scanning as a planned expansion.
Honest coverage today
Living Intelligence watches a curated starter set of 10 high-authority sources (of 17 registered); coverage expands as sources are reviewed — a human-reviewed process, not a continuous live web scan. Production coverage is not complete; it expands as more sources are reviewed and added.
AgentProof provides compliance-readiness and documentation support. It does not speak on behalf of any vendor, and it is not a certification, an audit, or regulatory approval. The full trust boundary is set out in the footer and on the Trust Centre.
Ready to build a compliance-readiness record for your own agent?
Request a compliance readiness pilot to apply this guidance to a real agent.
AgentProof builds a compliance-readiness record, not an official audit, and it does not speak on behalf of any vendor.