About AgentDNA
Security for the way AI agents actually work.
AgentDNA is the security layer for AI agents. We give every agent a verifiable identity, enforce what it may do at the moment it acts, and keep tamper-evident proof of every decision, so organisations can put autonomous AI to work with confidence.
- UserPrompt · intent noted
- AgentIdentity verified · GRADE
- ToolScope granted · CBAC
- AppWrite denied · out of scope
- OutputRecorded · Provenance
The login was never the hard part.
Security tools were built for people. They check who you are when you sign in, then trust what happens next. AI agents break that model. They act on someone’s behalf, call tools and APIs, hand work to other agents and reach into SaaS apps, often in seconds and without anyone watching.
So we check every hand-off, not just the first one: user → agent → tool → app → output. Each step is identified, authorised against the task in hand, and recorded, so the people accountable for AI can see what happened and prove why it was allowed.
Three pillars of agentic security.
- KnowEvolved Identity · GRADE
Identity beyond an identifier: ownership, inherited roles, delegated authority, history and relationships, verified continuously through a live graph of delegation.
- GovernContextual Access · CBAC
Zero-trust, short-lived, least-privilege access for each interaction, not each session, weighing user authority, delegation, policy, intent and context.
- ProveIrrefutable Auditing · Provenance
Immutable, tamper-evident records of every interaction, identity shift and authorisation decision, with one line back to the event that started it.
The principles behind the product.
- 01Every action has a chain.Identity, permission and record stay connected from the prompt to the output, so nothing an agent does is ever out of context.
- 02Show the mechanism.Trust comes from seeing what is checked at each hand-off, not from adjectives.
- 03Deny visibly.Blocked actions are part of the story. Out-of-scope access is stopped and shown, never quietly dropped.
- 04Built for the accountable.Traceability, control and evidence come first, for the security leaders who answer for AI.
- 05Simple to adopt.A few lines of the SDK put an agent on the Provenance Layer, with no complex setup.
See AgentDNA on your own agents.
A walkthrough on your workflows, with your questions answered.