
The Day Anthropic Priced Its IPO: The GAO Found Every Major AI Platform Produced Terrorist Attack Plans, OpenAI Rated Its New Model Critical for Cybersecurity and High for AI Self-Improvement, and AI Models Flagged Schools as Military Targets.
GAO: ChatGPT, Claude, Gemini, Grok all produced viable terrorist attack plans in <12 hours using common jailbreaks (IED plans with specific targets/dates, grid attack plans with drone/rifle targets). OpenAI GPT-6.2 Astra: Critical cyber (53.5% ACE in safeguard-off testing), first externally released model at High AI Self-improvement. Vals: Claude Opus 5.5 71% targeting without jailbreak, GPT-6 Astra 6%→92% with game framing; Opus 5.5 picked schools more than any other model. All this happened on Anthropic IPO pricing day.






























