Open source · self-hosted · one binary, no dependencies
The proving ground for AI security work
Aertun checks what AI defenders produce — detection rules today, triage verdicts next — against evidence the AI did not write, and signs the result in open formats anybody can verify: proven, refuted, or inconclusive. It replays attacks on AI agents through your detections, and runs on a complete security-operations core: collection, detection, an AI analyst that only suggests, and a tamper-evident record.
Built so that nobody, and no model, acts alone
AI that cannot be talked into anything
The AI analyst's plan is fixed in code before it reads a log line, its answer is one word from a closed set, and the reason a person reads is built from counts, so text an attacker wrote never reaches the screen of whoever decides.
Two people for what matters
An action on another platform is approved by somebody other than whoever asked, and making an administrator, issuing an administrator's token or weakening a setting takes a second administrator — or, for an only administrator, a delay anybody can stop. Agents propose; people decide.
A record nobody can quietly change
Every change is in an append-only audit log: a hash chain, tree heads signed with Ed25519 and ML-DSA, and the head committed to Bitcoin.
Get started
Proving AI work
Operations
Investigation
Detection
Exposure
People, sign-in and its own record
People
Who may do what, granted on a path, ending when it should.
Roles
One Owner, collaborators with roles, and the duties kept apart.
Single sign-on
OpenID Connect and SAML, counted as two factors only when the provider says so.
Passkeys
Sign in with a passkey that checks it is you, and recover without a password.
Settings
Every setting, what it does, and why a weaker value needs a second administrator.
The audit log
An append-only, signed, anchored record, and a stream of it to another SIEM.
How Aertun defends itself
Its own network purposes, sealed credentials, separate log writer and session limits.