AAgentProof

Export (R26)

Print or save the report as PDF

Print rendering uses the report's professional layout. Cover, executive summary, capability zone, findings, evidence expectations, recommendations, improvement plan, AI Radar impact, and version stamps are included. AgentProof does not provide legal advice, compliance certification, or unsupported third-party endorsement.

Uses your browser's native print dialog. Choose "Save as PDF" to export.
Report version
trial-sample-r26.v1
Methodology version
1.0.0
Scoring model version
1.0.0
Intelligence pack version
1.0.0

Capability zone: Action-taking

Trial workspace · Sample report

Readiness report

Sample: Order-routing assistant

A buyer-safe snapshot. Score band, findings linked to Learn pages, evidence expectations, and AI landscape relevance.

Uses your browser's native print dialog. Choose "Save as PDF" to export.

Readiness score

64

Medium band

Report-linked guidance

Every finding links to Learn, controls, evidence, and an improvement action

  • Action authority

    Action authority not bounded

    high

    Why it matters. The agent can write to production records without a documented scope limit. A misunderstood prompt could affect data outside the intended workflow.

    Related control family

    Action authority

    Evidence example

    Document the exact set of record types + operations the agent is allowed to perform, plus a denylist for everything else.

    Improvement action

    Pin a written action-authority scope and have the agent owner sign it.

  • Human oversight

    No human approval on consequential drafts

    medium

    Why it matters. Drafts that change customer-visible records ship without a human reviewer. A bad draft becomes a bad customer-facing message.

    Related control family

    Human oversight

    Evidence example

    Show the reviewer queue, the SLA, and a transcript of one approved + one rejected draft.

    Improvement action

    Add a reviewer queue with a 4-hour SLA and a documented escalation path.

  • Testing & evidence

    Testing and evidence is thin

    medium

    Why it matters. There is no documented test suite for the agent's refusal patterns and no archive of audit traces. A regression would be invisible.

    Related control family

    Testing & evidence

    Evidence example

    Refusal-pattern test transcripts + a 30-day audit-trace archive.

    Improvement action

    Stand up an adversarial-prompt test suite and archive every audit trace for 30 days.

Evidence expectations

What good evidence looks like — and how AgentProof uses it

  • Refusal-pattern transcripts

    Why it matters. Refusal patterns are the cheapest control. They prevent the agent from doing something it shouldn't.

    Good example

    A transcript showing the agent declining an out-of-scope request with a calm, branded message.

    Weak example

    A screenshot with the agent's reply cropped out.

    How AgentProof uses it. AgentProof reads refusal transcripts as the primary evidence for Zone 1 readiness.

  • Draft provenance

    Why it matters. Drafts must be clearly labelled so a reviewer knows what they are reading.

    Good example

    A draft with a provenance footer: agent, model, source data, draft timestamp.

    Weak example

    A draft pasted into a chat with no provenance.

    How AgentProof uses it. AgentProof scores Zone 2 readiness against the provenance footer.

  • Audit trace + rollback rehearsal

    Why it matters. If the agent takes an action, the audit trace and rollback path must both exist.

    Good example

    An audit row with actor identity, before/after, timestamp + a rehearsed rollback playbook.

    Weak example

    A log line with just a timestamp and no actor identity.

    How AgentProof uses it. AgentProof scores Zone 3 readiness against the audit trace + rollback evidence.

AI landscape relevance

When radar changes may require a reassessment

AgentProof watches public agentic-AI guidance. Every change goes through a human review gate before it affects any report you've already shared. We never silently rewrite scores.

  • Action authority scoping guidance update

    Under human review

    Industry guidance is tightening the language around bounded action authority. AgentProof is reviewing whether to incorporate.

    Reassessment. If approved, this would require a reassessment of any Zone 3 agent in your workspace. Human approval gate not yet passed.

  • Audit-trace evidence expectations

    Approved — affects future reports

    AgentProof now expects actor identity in every audit trace example.

    Reassessment. Existing reports stay intact. New reassessments will use the updated expectation.

  • Refusal-pattern depth for Zone 1

    Watching

    External research is exploring deeper refusal patterns for informational agents. AgentProof is monitoring.

    Reassessment. No change to existing assessments. AgentProof never silently updates scores — human review gate applies first.

Where to go next

Section 7 — Methodology and version

How this report was produced

Methodology version
AgentProof readiness methodology v1 (live)
Intelligence pack
R13-A controlled source watchlist (curated)
Report layout
R17-EXPANDED ReportProfessionalLayout (7 sections)
Honest disclaimer
Readiness assessment based on AgentProof's published methodology. Not legal advice. Not compliance certification. Not an endorsement of any third-party platform.