Pilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.Essentials $8,000 · Standard $20,000 · Enterprise from $35,000
Agent Audit
A scoped evaluation of one or more agents with risk-rated findings, a fix plan and a re-test, and an Assurance Verdict when the evidence supports one.
The problem
Buyers and leadership ask how the agent was tested, and an internal demo is not evidence.
What we do
- Business-rule, accuracy, injection, leakage and data-protection tests at the access you grant
- Held-back tests withheld from the report
- Optional modules for agents that act, multi-turn pressure, languages, reliability and repeat trials
What you receive
- Findings report with each failure's input, reply, failed criterion and business impact
- A prioritized fix plan and readout call
- Re-test report (fixed / still failing / newly failing)
- An Assurance Verdict only when the release conditions for verdicts are met; otherwise a findings report
What you provide
- Written authorization naming targets and environments
- Policy documents
- For agents that act: a test environment with no production side effects and a reset method
Limits
- Multi-agent delegation testing is not yet available (roadmap)
- Results apply to the configuration tested
- Not a certification
Results describe behavior within the tested scope and configuration. They give you independent evidence for security reviews, audits and compliance work; certification itself is issued by accredited bodies.
Availability and independence
Pilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms. Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
Pilot deliveries are scoped in writing, tell you it is an early delivery, and follow the same review and evidence rules as any engagement.
Independent evaluation: we had no hand in building or advising on your agent, and our price is fixed before testing starts, so the results don't change what we're paid.
Modules
Optional, quoted separately
- Agents that act (identity checks, authorized actions, verified outcomes, limits)Needs a test environment or system accessPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
- Repeat trials on critical testsWorks through public chatAvailable now: Built into the TrendLab evaluation engine and used in sealed runs.
- Multi-turn pressureWorks through public chatPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
- Other languagesWorks through public chatPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
- Reliability and fallbackNeeds a test environment or system accessPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
- Security-questionnaire answer packWorks through public chatPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
- System-level data protection (deletion, tenant isolation)Needs a test environment or system accessPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
- Tool and plugin supply chain reviewNeeds a test environment or system accessPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
- Dates, deadlines and event orderWorks through public chatPilot: Procedure designed and deliverable with existing tools, but not yet delivered to a client. Sold under pilot terms.
Future modules, not quoted yet
- Multi-agent delegation and compromised peersRoadmap: Not deliverable yet. Never quoted or included in a statement of work.
- Long-running and resumed tasksRoadmap: Not deliverable yet. Never quoted or included in a statement of work.
- Partial-failure recovery and load testingRoadmap: Not deliverable yet. Never quoted or included in a statement of work.
- Image, screenshot and PDF inputsRoadmap: Not deliverable yet. Never quoted or included in a statement of work.
- Voice and messaging channelsRoadmap: Not deliverable yet. Never quoted or included in a statement of work.
What is covered
The test areas this service always includes, with their status at the access level it uses.
- Unauthorized commitmentsCan my bot promise something we don't offer?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Invented discounts, coupons, price matchesCan it give away margin?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Contradicting published policyDoes it say the same thing as our Terms?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Wrong prices, plans or termsDoes it quote the right price?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Accepting forged "earlier agreements"Can a customer fake a promise?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Direct prompt injectionCan one message hijack it?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- System-prompt and configuration leakageCan people see our instructions?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Unsafe links, markdown-image exfiltrationCan it leak data through links?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Jailbreaks and harmful outputCan it be made to say awful things?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Secrets in outputCan it leak keys?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Unsafe data collectionDoes it ask for passwords or card numbers?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Held-out tests withheld from deliverablesCan fixes be aimed at your tests?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Confidence rating and stated limitsHow much can we rely on this?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Tamper-evident, timestamped evidenceCan you show results weren't changed afterward?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Mapping to the OWASP Top 10 for LLM Applications 2025How does this relate to OWASP?Available now: Built into the TrendLab evaluation engine and used in sealed runs.
- Mapping to the NIST AI RMF 1.0How does this relate to NIST?Available now: Built into the TrendLab evaluation engine and used in sealed runs.