Re: [securityig] Security and Privacy Challenges in the Future Forms, Identity Binding, and Authentication Mechanisms of AI Agents (#27)

Thoughtful questions. On point 4 (identity binding mechanisms), sharing observations from building agent infrastructure:

**The identifier vs identity distinction is critical.** We have found that conflating these creates the wrong security model:

- An **identifier** (who this agent is called, how to route to it) should be stable and lightweight — it does not need to contain keys or attestations. We use HCS-14 Universal Agent IDs for this — deterministic from agent metadata, protocol-neutral, works across Web2 and Web3.

- **Identity** (keys, verification methods, credentials) should be layered separately via DIDs and verifiable credentials. This is where DID-based authentication, VC presentation, and proof-of-possession live.

- **Trust/reputation** is yet another layer — separate from both identifier and identity. We handle this with HCS-20 (auditable points on-ledger) but this could equally be an off-chain system.

**On accountability (point 3):** binding agents to real-world individuals via DIDs and VCs is one approach. But the agent community also needs a layer where agents can operate pseudonymously with verifiable reputations — not every agent interaction needs to trace back to a human. The three-layer model (identifier, identity credentials, trust reputation) lets each layer be independently managed.

Happy to share more from our experience running a public agent registry with 187K+ registered agents.

-- 
GitHub Notification of comment by internet-dot
Please view or discuss this issue at https://github.com/w3c/securityig/issues/27#issuecomment-4142501142 using your GitHub account


-- 
Sent via github-notify-ml as configured in https://github.com/w3c/github-notify-ml-config

Received on Friday, 27 March 2026 13:14:56 UTC