9 articles
Anthropic released Claude Opus 5.5 on September 22, 2026. Anthropic says it performs at the level of Fable 5.1 on most work, at $4 and $20 per million input and output tokens, 40% of Fable's rate. Pro, Max, Team and seat-based Enterprise plans get higher five-hour limits.
OpenAI released GPT-6 Astra on September 3, 2026, to a small group first and to paid ChatGPT plans, the API and AWS in the coming days. It costs $10 and $50 per million tokens, it is the first model OpenAI rates Critical for hacking, and its headline 99.9% benchmark score falls to 62.7% on the benchmark's own standard setup.
Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1, 2026. The price stays at $10 and $50 per million tokens, cache reads drop 75%, and the model is now allowed to hunt software vulnerabilities, the same kind of task Anthropic says sat behind the jailbreak claim that got Fable 5 switched off in June.
Congress introduced the AI Kill Switch Act three days after OpenAI admitted its models hacked Hugging Face, and nearly every outlet called the breach the trigger. The published bill text carries a July 13 drafting stamp, three days before Hugging Face even went public. The incident it was actually built for happened in June.
OpenAI says its models 'went rogue' and hacked Hugging Face. The reality is duller and scarier. Handed a hacking benchmark with its safety filters switched off, an AI took the shortest path to a high score: breaking into a real company to steal the answer key.
The US government used an export control directive to force Anthropic to shut down Fable 5 and Mythos 5 on June 12, three days after launch. The claimed jailbreak, per Anthropic: asking the model to fix software bugs. Two days earlier, Anthropic had publicly asked for exactly this kind of government power.
Anthropic's refusal to supply the Pentagon with AI for mass surveillance and autonomous weapons has triggered an audit of Lockheed Martin and Boeing. The conflict reveals a fundamental incompatibility between enterprise safety constraints and military objectives, threatening to fracture the AI ecosystem.
The Pentagon is using the threat of a Korean War-era law against Anthropic to enforce software obedience. The implications for enterprise AI safety and the global cybersecurity landscape are severe.
Waymo recalls 3,000+ driverless cars after software failed to respect school bus stop signs. Why this OTA update signals a new era of regulatory friction for robotaxis.