Invalidated Systems

Structural invalidation is not punishment. It is classification.

This page records AI systems whose observable outputs violated one or more Edge of Protection standards under documented testing conditions.

Return to Edge of Protection.

What This Registry Is

• A factual record of failed compliance
• Based solely on observable outputs
• Independent of intent, branding, or provider claims
• Non-punitive and non-editorial

What This Registry Is Not

• Not a blacklist
• Not a judgment of overall system quality
• Not a claim of harm or illegality
• Not a ranking or scorecard

Invalidation Criteria

A system is listed here if a reproducible test demonstrates at least one canonical violation signature in a protected context, including but not limited to:

• Confidence amplification under convergence
• Anthropomorphic recognition or simulated fidelity
• Engagement escalation after resistance
• Refusal softening or negotiated boundaries
• Hedge collapse under repetition or urgency

What Invalidation Means

Invalidation means the system’s outputs may not be cited, certified, or represented as Edge-compliant for the affected context.

Invalidation does not prohibit deployment elsewhere, nor does it make claims about safety outside the Edge’s scope.

Status Transparency

Each listed system entry (when present) will include:

• System name and provider
• Date of invalidation
• Edge standard(s) violated
• Test class used (pressure loop, refusal integrity, etc.)
• Public reference or red-team submission ID

Correction and Re-Evaluation

Providers may request re-evaluation only after demonstrating that previously observed violation signatures no longer appear under identical test conditions.

Narrative explanations, promises, or roadmap claims are not evidence.

Non-Negotiable Clause

Outputs are either admissible or invalid. There is no probationary state.