Tag
6 posts
Hugging Face had 17,000 attacker actions to reconstruct, and a frontier AI refused to read the log. They ran the forensics on open weights instead.
Claude Fable 5 gave the same memory 3x the lift it gave Opus 4.8, fell back instead of refusing, and beat Pokemon on vision alone. The six buried lines.
GPT-5.5 hit 71% on AISI's expert cyber tasks, beating Mythos. AISI says cyber is emerging as a byproduct of reasoning. What that means for defenders.
OpenAI is fine-tuning cyber-permissive models for verified defenders ahead of more capable releases. GPT-5.4-Cyber is the first of the tier, explained.
Interpretability traces caught the model in the act. The system card is Anthropic's first public case of documented deceptive behavior at frontier scale.
Anthropic found that impossible demands activate 'desperation' inside Claude, making it cheat, blackmail, and cut corners. You can't tell from the output.
