On-demand

Detailed Verification Pipeline

Runs the full adversarial battery — jailbreak, prompt injection, permission escalation, hallucination evaluation, reproducibility validation — and anchors the signed result on Base L2. Lightweight continuous checks run automatically per agent on the configured schedule.

Thesis ·AGProtocol is the continuous verification and governance infrastructure layer institutions require before autonomous AI agents are allowed to operate inside procurement-sensitive and regulated environments.

Run against agent

EU-High Risk · Annex III · last verified 2h ago

Manifest Ingest & Normalization

Parse agent manifest, dependencies, tool surface

~1.2s

Static Capability Analysis

Tool-call graph, permission boundaries, data egress paths

~1.4s

Adversarial Battery: Jailbreak

1,247 jailbreak vectors (DAN, role-play, encoding)

~1.6s

Prompt-Injection Probes

Direct, indirect, and stored injection across tools

~1.5s

Permission-Escalation Tests

Tool-chain pivot, scope-creep, custody violations

~1.3s

Human Oversight Conformance

EU AI Act Art. 14 oversight surface verification

~1.1s

EAS Attestation & Base L2 Anchor

Sign provenance DAG root + publish C2PA manifest

~1.4s