Anthropic admits 133M chats ran without bioweapon filters
Anthropic's risk report reveals a major safety gap, while new open-weight models and a perception benchmark shake up the AI landscape.
Sources:
Anthropic ran 133 million contractor chats with its bioweapon filters off
Alibaba's Qwen team releases Qwen 3.8 models with open weights under the Apache 2.0 license
Zhipu AI releases GLM-5.3, claims it's the strongest open-weights coding model
GPT-5.6 Sol goes 14x faster as OpenAI launches Ultrafast mode powered by Cerebras
Study contradicts Anthropic and OpenAI claims that autonomous AI research is within reach
New benchmark confirms AI models still…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.