Daily AI Roundups
Anthropic launches free, opt-in OSS Scanner for eligible open-source projects. OpenAI rolls out GPT-6.1 Sol Ultrafast at up to 8x Standard speed. FAR.AI reports 98.8% AUC for deception probes in a preprint. Google previews its enterprise Gemini agent. Hugging Face releases TRL v1.15.0 for longer training sequences in single-GPU tests. Tencent publishes its 5B-parameter Youtu-Parsing-Omni model. StepFun’s Step 5 Preview reaches OpenRouter. Arena previews an Alignment Index across 27 models. Philip Zeyliger demonstrates shared-terminal control with Shoulder. Geoffrey Huntley argues for better CLI help. The White House announces $2.4B in Genesis Mission tools and compute credits. GlobalFoundries signs a $2 billion TSMC interposer deal. Three fired OpenAI safety researchers publish a misconduct denial. USA Today seeks more than $250 million in its OpenAI copyright lawsuit.
- AI security
- AI models
- AI agents
- Developer tools
- AI research
- AI policy
Stories in this roundup
- Three fired OpenAI safety researchers deny misconduct in a newly published letterTomek Korbak, Jasmine Wang and Mikita Balesni