Sovereign security AI: Mythos-class defense on your own GPUs (Chitos)

Sovereign security AI: Mythos-class defense on your own GPUs (Chitos)

Cloud attack-security AI (e.g. Anthropic Mythos) is powerful but cloud-only — your code leaves your building to use it. For defense/gov/finance/healthcare, that is a non-starter.

Chitos is a sovereign, on-prem autonomous security-AI harness: find it before attackers do, then prove it — with no data egress.

Two engines

• Deterministic static analysis — reproducible detection (injections, deserialization, SSRF/XXE/XSS, credential exposure, memory safety, auth, logic), CWE-mapped, no AI needed (0 cost, 0 egress).
• AI reasoning layer — verifies + explains findings, reasons kill chains. => dual evidence vs one black-box verdict.

Trust features

• Metacognitive self-check: before asserting a vuln, it asks false positive? reproduces? real source to sink flow? -> structurally fewer false alarms.
• CodeGraph grounding: maps definitions/deps/risky call sites to localize analysis.
• 3 modes: Scan / Research (CVE web lookup) / Exploit (kill chain on approved target). Code snippet or URL input.
• Default: DeepSeek V4 Flash (304B MoE, MIT) on your infra; Claude via your key optional.

Honest scope

Mythos is stronger at raw discovery — Chitos does not claim otherwise. Chitos wins on sovereignty, transparency, reproducibility, regulatory fit, cost. Mythos-class defense, on your own GPUs, no egress.

Free, no signup: https://chitos.vidraft.net