Why AI agents need a trust layer

Identity, a runtime gate, human approval and signed evidence between an agent's decision and the action it takes

Companies built identity for people and logging for applications. Neither was built for software that decides what to do next and then does it with a company card, a customer's refund or a production database. This paper defines the agent trust layer, the checkpoint between an agent's decision and its action, and shows how Vorim builds it: a key for every agent, a deterministic gate outside the model, approval routed to named people and rationed, and signed evidence anyone can verify offline.

• Why identity products, logs and model guardrails each miss the moment an agent acts
• The five questions every consequential agent action raises, and what answers each one
• Four worked decision records: a refund, a prompt injection, a looping swarm and an off-intent action
• How intent, drift scores and swarm charters turn behaviour into something you can measure
• A plain table of what's shipped and what isn't, including the three things we haven't built

1. Executive summary
2. Agents act, and their authority has no home
3. Why the existing layers cannot close the gap
4. The category is the agent trust layer
5. How Vorim works, as an operating model
6. The technical foundation
7. Decision records, intents and swarms
8. Evidence, standards and deployment
9. Where it pays off
10. Why now
11. The call to category
12. Design partner programme

20 pages, 35 min read. Published for Cybersecurity Awareness Month, October 2026. May be shared and quoted with attribution.