The first release of Aertun, the proving ground for AI security work: it holds what AI defenders produce to evidence the AI did not write, says proven, refuted or inconclusive, and signs that — in one binary with no dependencies.

Watch the tour (eight minutes, captioned): https://aertun.github.io/#tour

What it proves

Detection rules against a labelled corpus · your AI analyst, asked repeatedly with instructions planted in log text · the analyst's auto-close threshold · another vendor's AI verdicts against your people's decisions · the same on a blind random sample · a vendor's pre-registered claim · an agent's own account against what was seen · a VEX document's "not affected" statements · every citation in an AI's investigation. Each proof is an in-toto statement in a DSSE envelope, signed with Ed25519 and ML-DSA-65, and checks without Aertun (scripts/verifyproof/verify.py).

Verified for this release

• Live in Docker: every command-line step, every console page in Chrome (no console errors, no accessibility issues, light and dark, phone width), HTTP and MCP security checks, OpenID Connect and SAML against Keycloak.
• SigmaHQ: all 4,544 rule files read as PyYAML reads them; 3,076 of 3,164 core rules compile.
• govulncheck: no reachable vulnerability on Go 1.27.4.

The full notes are in NEWS; measurements and method are in the README.

Check what you download

Every artefact carries its SHA-256 in SHA256SUMS, and each has a build provenance attestation:

sha256sum -c --ignore-missing SHA256SUMS
gh attestation verify aertun-linux-amd64 --repo aertun/aertun

The container image is ghcr.io/aertun/aertun:0.1.0 (also 0.1 and latest): the binary alone on distroless, no shell.