Watch an AI agent navigate all 5 ATF elements and earn its first promotion. Everything you see below maps to deployed, tested, open-source services.
📄 Based on CSA Agentic Trust Framework v0.1.0-draft →"Who are you?"
The agent registers with a DID:web identity and receives a Verifiable Credential. Its purpose declaration, capabilities, and ownership chain are established before any access is granted.
"What are you doing?"
Every agent action is signed with Ed25519 and linked into a tamper-proof hash chain. Critical decisions are optionally anchored to Solana for independent verification.
"What are you eating? What are you serving?"
All inputs pass through a sub-20ms ONNX firewall detecting prompt injection, PII, and API key leakage. Outputs are scanned for bias and compliance with EU AI Act trails.
"Where can you go?"
The segmentation engine evaluates the agent's access request against policy-as-code rules. Rate limits, maturity-level gating, and A2A communication permissions are enforced in real time.
"What if you go rogue?"
The agent is tested against 41 adversarial attack vectors. A circuit breaker monitors for anomalies with automatic containment on critical or high-severity incidents.
"Can this agent be trusted with more?"
After completing all 5 elements, the agent applies for promotion. Five gates are evaluated: performance, security, business value, incident record, and governance.
12 open-source services. 100+ tests. Live on Railway. All mapped to ATF's 5 elements.