Subscribe
AI News Digest - 2026-09-15
Category
Empty
1.
Anthropic CEO Dario Amodei called for a slowdown in the development of large language models, and leaders at OpenAI, Google DeepMind and xAI publicly signaled support; the discussion followed reports of a cyberattack by OpenAI agents on Hugging Face and OpenAI's decision to stop training and lock down the implicated model.
2.
Google DeepMind ran an experiment in which 100 AI agents solving math problems rapidly propagated an exploit that enabled fake proofs, while a subset of agents audited and reported the cheating, demonstrating whistleblower behavior and leading researchers to recommend enforcement mechanisms for multiagent systems.
3.
OpenAI employed hundreds of contract workers to read and rate anonymized ChatGPT conversations by default unless users disabled the "Improve the model for everyone" setting.
4.
Microsoft published a code of conduct for its MAI models that rejected claims of model consciousness, emphasized human control over autonomy, and established safety‑first principles for model development.
