Tripwire Status
Tripped (Mandatory Stop)
Peak Capability Score
92.4
/ 100
Time to Threshold
18
days
Societal Risk Level
Contained
Continuous Capability Trajectory vs. Containment Margin
AI Capability
Safety Margin
Kill Switch
Simulating Day 0–45 recursive self-improvement dynamics. Automated shutdown clamps compute cluster before unconstrained autonomous runaways.
Model: Latency × Autonomy
Context & Verified Reporting Grounding:
BBC Interview • Jack Clark (Anthropic Co-founder)
Anthropic co-founder Jack Clark told the BBC that frontier AI systems are becoming more powerful “by the day”, warning that artificial intelligence could pose a “major risk to society” without hard regulatory circuit breakers. In catastrophic risk governance, mandatory hardware and datacenter kill switches ensure an automated power-down or weight scram is executed whenever autonomy, self-exfiltration, or deceptive self-improvement metrics cross verified threat thresholds.