In a week of significant developments, Anthropic and the US government have agreed to lift restrictions on the company, allowing it to redeploy Claude Fable 5 after a period of hiatus. The agreement includes the addition of new cybersecurity classifiers, a draft jailbreak-severity framework, and expanded model-testing coordination. However, broader concerns remain about the inevitability of jailbreaks and uneven release constraints compared to competitors like OpenAI.
New Models and Tools
Anthropic also launched Claude Sonnet 5 with time-limited discounted pricing, improved agentic coding and benchmark performance, reduced misaligned behavior, and default cyber safeguards. The launch comes despite Sonnet 5 having relatively weaker cybersecurity capability than top-tier models.
In other tools and app updates, Google NotebookLM is now generating TikTok-style vertical video summaries of uploaded research. Google also released Nano Banana 2 Lite, a faster, cheaper image generator available via API.
Hardware and Business Updates
Business and research updates are also making waves. AI hardware startup Etched is pulling 400+ engineers from NVIDIA, TSMC, and more to build a new frontier inference cluster for AI, which is already worth $1 billion in demand. Meanwhile, Baidu's AI chip unit is eyeing an IPO, Agility Robotics is planning a SPAC, and DeepSeek is expanding its hiring.
The week also saw news about China's open-source Longcat 2.0 MoE model, which features notable large-scale training and efficiency techniques alongside new long-horizon agent benchmarks. In other business news, Neon bought the film "Artificial," about OpenAI, after Amazon dropped it, and Tidal updated its policy to not pay royalties on AI-generated music, though it is not banning it outright.