Research program / AI adversarial testing
Demon Core
What happens when a consequential AI system is pushed until a failure stops being isolated?
Demon Core explores authorized high-intensity testing, propagation tracing, containment, evidence capture, and remediation replay. Public copy deliberately stops short of claiming production readiness or deployed performance.
Boundary: the public experience describes the research direction and campaign model. Its telemetry is interface storytelling, not a live test.
Controlled pilot / AI agent qualification
GhostGate
Can this exact AI-agent version be admitted to this environment?
GhostGate registers the exact version, qualifies behavior against agreed scenarios and policy, preserves evidence, routes the decision to human reviewers, and supports signed release attestations.
Evidence path: controlled and deterministic proof is published separately from synthetic scale. Customer production validation is not claimed.
Private-pilot evaluation / Cyber-extortion evidence
Proofline
Did the attacker prove the claim—or only provide material that appears persuasive?
Proofline gives response teams a local-first workspace for comparing claims, supplied artifacts, authorized reference information, conflicts, gaps, and reviewer findings.
Boundary: Proofline is being prepared for limited private-pilot evaluations. It is not an automated verdict engine, attribution system, negotiator, or substitute for forensic and legal disciplines.
Research preview / AI-memory incident recovery
Recall
What did known contaminated knowledge affect, and how can candidate repair paths be examined safely?
Recall explores propagation across retrieval, decisions, behavior, and artifacts, then uses bounded counterfactual replay to evaluate candidate recovery paths.
Boundary: production remediation remains human-authorized. Policy-scoped automation is planned only for reversible, high-confidence actions; it is not presented here as a current production capability.