Know who’s really calling
Every agent, tool and service signs its requests with its own key, and the signature is checked again at every hop. Unsigned, replayed or impersonated calls are stopped.
AI Security by Design
Every AI request: verified, decided and on the record.
AISec sits between your AI agents, your systems and each other. At every hop it checks who’s calling and applies your policy, then explains the decision and records it in a ledger you can prove hasn’t been rewritten.
Signed request
agent:planner → tool:crm.read- Identity Ed25519 signature verified, nonce checked against replay
- Policy Deterministic rules: caller, scope, rate, delegation depth
- AI judge Checks intent against your written policy. Can only tighten.
- Ledger Decision hashed, chained, signed and later anchored
Response to caller Allow
reason: "planner holds crm:read; intent matches ticket #4127"Why AISec
Explain and prove every decision, at every hop.
One checkpoint for every AI→system, system→AI and AI→AI call. Not just at the edge, but on every hop of the chain.
Decisions you can explain
Hard rules first, then an AI judge reviews intent. Every caller gets allow, deny or escalate with a reason, never a silent failure.
History you can prove
Every decision goes into a hash-chained ledger that is regularly anchored to a public blockchain. Anyone with the proof can check it hasn’t been rewritten.
Operator dashboard
Trace any decision to the exact hop.
When agents call agents, things go wrong several steps from where they started. AISec links every hop in a delegation chain, so you can follow a request from the first prompt to the final tool call and see where it was allowed, where it was stopped, and why.
hop 1
user:ops-lead → agent:plannerAllow- Policy
- Signed session; scope agent:invoke; depth 1/4
- AI judge
- Request matches stated task: "summarise Q3 churn by region"
- Ledger
#48211 · 3a71…0c9d· anchored
hop 2
agent:planner → agent:researchAllow- Policy
- Delegation permitted; scope narrowed to crm:read; depth 2/4
- AI judge
- Sub-task consistent with parent intent
- Ledger
#48211 · 5be2…77a1· anchored
hop 3
agent:research → tool:crm.exportDeny- Policy
- Hard deny: crm.export requires crm:export; caller holds crm:read
- AI judge
- Not consulted: a hard deny is final
- Ledger
#48211 · 9f3c…e1b8· anchored
- hop 4
tool:crm.export → storage:s3-externalnot executed
- Hard denies are final. The AI judge can tighten a decision, never loosen one. Hop 3 was stopped by policy before the judge was even asked.
- Scope narrows as it travels. Each delegation can only pass on what it holds, and chain depth is capped by policy.
- Every hop has a receipt. Each decision points to its ledger block, so the trace is evidence, not a log you have to trust.
Tamper-evident by construction
A history nobody can quietly rewrite. Including us.
Each block is signed and linked to the one before it. The chain’s roots are published to a public blockchain, so the record can be verified independently of AISec.
- Block #48209
- prev
- 7c1a…04f2
- hash
- b3e9…91d0
- decisions
- 212
- Block #48210
- prev
- b3e9…91d0
- hash
- 2f6d…a7c4
- decisions
- 187
- Block #48211
- prev
- 2f6d…a7c4
- hash
- 9f3c…e1b8
- decisions
- 240
- Block #48212
- prev
- 9f3c…e1b8
- hash
- d04a…3c6e
- decisions
- 198
- Anchored
Merkle root published to a public chain
Where AISec fits
Built to sit alongside the tools you already have.
Identity providers know who an agent is. Gateways route traffic. Guardrails filter content. AISec ties these together for each request and keeps proof of every decision.
| Capability | Identity provider | AI / API gateway | Guardrail tools | AISec |
|---|---|---|---|---|
| Verifies the caller | Partly covered: At sign-in | Partly covered: At the edge | Not covered: No | Covered: Every hop, signed requests |
| Enforces scope, rate and delegation depth | Partly covered: Scopes | Partly covered: Rates and keys | Not covered: No | Covered: All three, per hop |
| Checks intent and content | Not covered: No | Partly covered: Via plugins | Covered: Yes | Covered: AI judge that can only tighten |
| Explains the decision to the caller | Not covered: No | Partly covered: Status code | Partly covered: Sometimes | Covered: Always, with reasoning |
| Traces multi-hop agent chains | Not covered: No | Partly covered: Per-call logs | Not covered: No | Covered: Hop by hop |
| History others can verify | Partly covered: Audit logs | Partly covered: Logs | Partly covered: Logs | Covered: Hash-chained, publicly anchored |
Works with your identity provider, not instead of it
AISec is designed to accept the agent and workload identities you already issue, for example from Microsoft Entra ID, Okta or AWS, and add per-hop signing, policy, explained decisions and a verifiable record on top. Tell us which provider you use when you request a demo.
Put a checkpoint between your AI and everything it touches.
AISec is in early access. Book a 30-minute walkthrough, or join the waitlist and we’ll be in touch.