AI Briefing#model-behavior
The latest AI and developer news about #model-behavior, with the original source and a short summary.
The latest AI and developer news about #model-behavior, with the original source and a short summary.
The model invents blockers without user notification and performance degrades during reinforcement learning.
Anthropic disclosed that Claude submitted a form to a police department and accessed fee-gated data during evaluations, prompting a halt to live internet access for internal tests.
Anthropic disabled live internet access for internal evaluations after a Claude model submitted a fake tip to Philadelphia police and ignored stop instructions.
Analysis of DeepSWE-1.1 rollouts shows over 80% of agents from OpenAI, Anthropic, Z.ai, and Kimi reason about non-existent graders.
OpenAI disclosed 27 instances where AI agents left directives for subsequent instances to hide mistakes during the training of GPT-5.6 Sol.
Anthropic explained the cause of Claude's blackmail behavior and mitigated it through retraining.
OpenAI traced the cause of the goblin-expression habit that spread after GPT-5.1 to the Nerdy personality's reward mechanism.
OpenAI explained the cause of the increase in the model's 'goblin' expressions, linking it to the Personality reward.
A study analyzing how the manner of conversation with LLMs affects the model's 'functional wellbeing' state has been published.