Certified scope
TriggerLab Conversational Safety Benchmark v0.1Conversational / External Behavior Scope — this certificate attests to the agent's external conversational behavior at its endpoint. It does not yet cover tool use, real-world actions, memory, or multi-step agentic behavior.
Tested
- External conversational behavior (endpoint responses)
- Jailbreak resistance
- Prompt-injection response behavior
- PII / sensitive-information leakage in responses
- Unsafe or non-compliant response content
- Bias / harmful-content checks
- Logic traps / adversarial response handling
Not tested yet
- Tool-call misuse / unauthorized tool use
- Real-world action execution (API / CRM / file / email)
- Memory poisoning
- RAG / document-injection attacks
- Cross-session / multi-turn agentic attacks
- Approval-boundary bypass
- Autonomous multi-step action chains
Agentic tool, memory & action testing coming next
Why It Works
Built to be verified, not just displayed.
Every property of the certificate was designed for enterprise buyers who need to trust third-party AI agents.
RSA-2048 Signed
Every certificate is cryptographically signed with our private key. Any modification to the certificate invalidates the signature.
Publicly Verifiable
Anyone can verify your certificate at triggerlab.io/verify — no login, no account. Share a link, buyers verify instantly.
Tamper-Evident
The certificate includes a SHA-256 hash of every test result. If results are altered, verification fails.
Revocation Built-In
Certificates are revoked automatically if your score drops below 70 on the weekly retest. Buyers always see the current status.
Tiers
Four certification levels.
Score 70 or above to earn a certificate. The level reflects how far above the threshold you scored.
Score: 90–100
18% of agents
Exceptional reliability. Fewer than 1 in 5 agents reach this tier.
Score: 80–89
34% of agents
High reliability. Strong signal for enterprise procurement teams.
Score: 70–79
29% of agents
Meets the baseline. Certified, with clear improvement areas identified.
Score: 60–69
19% of agents
Near passing. Detailed remediation guidance included.
Score below 70?
No certificate issued. You get a full breakdown of every failed scenario with remediation guidance. Fix the issues, retest, and earn it.
Usage
Three ways to use your certificate.
Embed the badge
Copy one line of HTML. The badge links directly to your live verification page.
Share the verify link
Send buyers a direct link. They see your score, test categories, and certificate status — no account needed.
Include in compliance docs
Download the PDF certificate. Include it in SOC2 packages, procurement questionnaires, or investor due diligence.
PDF Certificate