The Trump administration has lifted restrictions on Anthropic's Mythos and Fable AI models, clearing a regulatory path for the AI company to deploy enhanced versions of its technology. Anthropic is redploying Claude Fable 5 with new cybersecurity classifiers and is collaborating with major industry partners to establish a jailbreak-severity framework, signaling coordinated efforts to address AI safety concerns across the sector. Concurrent with the policy shift, Anthropic unveiled Claude Sonnet 5, positioning the model as a cost-effective option for building agentic AI applications. The company is offering time-limited discounted pricing and highlighting improvements in coding performance, benchmark scores, and reduced misaligned behavior. Despite having weaker cybersecurity capabilities compared to larger models, Sonnet 5 includes default cyber safeguards as standard, reflecting industry emphasis on safety guardrails. The week saw substantial activity across AI infrastructure and corporate developments. Etched is raising major funding to build specialized inference hardware, recruiting experienced engineers from NVIDIA and TSMC. In parallel, Baidu is pursuing an IPO for its AI chip unit, DeepSeek announced plans to double its workforce, and China released LongCat 2.0, an open-source mixture-of-experts model featuring advanced efficiency techniques. Google also released new tools including NotebookLM video summaries and Nano Banana 2 Lite, a faster, cheaper image generator.