TitusOS Citadel

Citadel is the autonomous defense system that monitors and protects every TitusOS workspace. Concentric defense-in-depth, founder-in-the-loop on anything customer-impacting, and a tamper-evident audit chain over every decision.

How Citadel works

Citadel is structured as a fortress: every component has a single, named job, and they reinforce each other. Perception, decision, response, and verification are separate layers — a single compromise can't cascade.

Founder-in-the-loop

Citadel acts on its own only for low-risk, high-confidence responses (rate-limit a noisy IP, revoke a session, quarantine a token). The <30s containment target applies to that set. For anything customer-impacting, a founder must approve.

• High-impact approval gate. Founder approval required before Citadel locks accounts, disables tenants, mass-rotates credentials, or takes a service offline.
• War Room review queue. Every Citadel action — even autonomous low-risk responses — lands in a queue for after-the-fact founder sign-off.
• Novel-threat escalation. Citadel only acts autonomously when confidence is high and the playbook has fired successfully before. Anything novel pages a human and waits.
• Training Yard sign-off. Self-attack findings require founder triage before Citadel auto-patches or files tickets — a self-modifying defense system without a human checkpoint is a liability, not an asset.

Public commitments

• Sub-30-second containment for low-risk autonomous responses
• Founder-in-the-loop on every customer-impacting action
• Tamper-evident audit chain over every Citadel decision
• Quarterly transparency report — what we saw, what we caught, what we missed