Anthropic changed Fable 5's biology safeguards, sending fewer ordinary biology requests to a less capable model while keeping dual-use restrictions in place. OpenAI expanded free ChatGPT access to GPT-5.6 Luna and updated GPT-5.6 Sol for Plus and Pro users, while Hugging Face added Baseten as a hosted text-model provider. These changes gave free users unlimited text chats and gave developers another route to hosted language models. Google DeepMind's reported AI system for forecasting cyclones remained unconfirmed by peer-reviewed or independent validation.
Text+5Aug 7
Image: anthropic.com
Anthropic narrows Fable 5 biology fallbacks
Everyday biology and health questions should face fewer handoffs to a less capable fallback model, while dual-use restrictions remain in place. Anthropic changed Fable 5's biology safeguards and said biology-related fallbacks fell by about 85% in its testing. Healthcare professionals may receive more help with clinical tasks.
What still breaks: Fable 5 still sends dual-use requests to Opus 5 and is not yet usable for professional biology research or drug development.
Text+2Aug 6
Image: openai.com
OpenAI updates GPT-5.6 Sol and free Luna access
Free ChatGPT users can now use GPT-5.6 Luna as their default model for unlimited text chats. Plus and Pro users received a GPT-5.6 Sol update that OpenAI said is more factually reliable and focused.
What’s still unclear: The size of the factual-reliability gain has not been independently established.
Text+2Aug 6
Hugging Face adds Baseten as a text-model provider
Developers can now use Baseten-hosted language models through Hugging Face tools and billing. Baseten joined Hugging Face Inference Providers and is available through its Python and JavaScript software-development kits, which are code libraries for developers. The initial support covers conversational and text-generation models.
Worth knowing: The initial integration covers conversational and text-generation tasks, with additional task types described only as a future rollout.
OtherAug 6
OpenAI and APA begin youth mental-health collaboration
The collaboration could inform future youth-safety practices. It gives users no new safety control or product behavior to rely on. OpenAI said it is working with the American Psychological Association to bring psychological science into responsible-AI development and youth mental health, without announcing a product feature, safeguard deployment, or policy taking effect.
Not scored - A partnership and convening do not themselves deploy a safety control or change ChatGPT; collaboration language alone does not move the risk dial.
Announced or reported, but nothing you can use or verify. Some of these will become real. Some won’t.
