Monday, 27 July 2026 Sign in

Last Week in AI #251 - Mythos Back, Sonnet 5, Etched, LongCat

Last Week in AI Last Week in AI 21 July 2026, 11:31 ● Covered by 4 sources

The Last Week in AI podcast episode 251 covers Anthropic's redeployment of Claude models after US government talks with new cybersecurity features, the launch of Claude Sonnet 5 with improved agentic capabilities, and updates from Etched, DeepSeek, and China's LongCat 2.0 open-source model. Etched is building specialized inference hardware with backing worth $1 billion in customer demand, while Agility Robotics plans a $2.5 billion SPAC merger. The episode also discusses new benchmarks for agent systems, Google's video generation tools, and policy developments including Taiwan's Nvidia smuggling investigation.

Why it matters

Trump lifts restrictions on Anthropic, Anthropic launches Claude Sonnet 5, Google's NotebookLM updates, chips stories from Etched and Baidu, and more!

Read the full story at Last Week in AI

Also covered by

AI Agents Anthropic

Related stories

LWiAI Podcast #235 - Sonnet 4.6, Deep-thinking tokens, Anthropic vs Pentagon Last Week in AI · 4 months ago

Deep Learning Weekly: Issue 459 Deep Learning Weekly · 1 month ago

LWiAI Podcast #249 - Fable 5 ban, SpaceX Cursor + IPO, OSS Aplenty Last Week in AI · 5 days ago

LWiAI Podcast #234 - Opus 4.6, GPT-5.3-Codex, Seedance 2.0, GLM-5 Last Week in AI · 5 months ago

Deep Learning Weekly: Issue 462 Deep Learning Weekly · 3 weeks ago

Deep Learning Weekly: Issue 445 Deep Learning Weekly · 4 months ago

The daily briefing

Every AI story that matters,
in your inbox by 8am.

TLDRocket reads 60+ sources, removes duplicate coverage, and summarises the day in two minutes. Free, no spam, unsubscribe anytime.

TLDRocket 26 July 2026

The AI ecosystem is fracturing along three fault lines: geopolitical protectionism masquerading as security, architectural shifts that reshape where value accumulates, and the messy reality of how AI systems actually get built and exploited. Start with China. Moonshot AI's Kimi launch triggered the familiar cycle—American frontier AI companies lobbying regulators to restrict open Chinese models, citing national security. But as reporting shows, these concerns often conflate legitimate questions about AI safety with protectionism that primarily benefits OpenAI and Anthropic. The irony cuts deeper: while regulators debate Chinese open weights, a thriving shadow market has emerged in China where resellers bundle stolen API credentials and exploit free trials to undercut official pricing, exposing how porous these systems actually are. Meanwhile, the technical frontier is tilting toward multimodal unification and architectural simplification. Black Forest Labs' FLUX 3 generates images, video, audio, and robot actions from a single unified model, while the Model Context Protocol ditches stateful sessions for stateless HTTP-like services, letting developers deploy without specialized machinery. These changes sound procedural but they're about efficiency—reducing friction in how AI gets built and deployed. The governance fight between Microsoft and Google DeepMind reveals where the real competition is moving. Nadella wants enterprises to own their data and switch models freely, routing orchestration value through Azure. Hassabis proposes an industry-funded regulatory gate testing frontier models before release—advantaging incumbents like Google DeepMind that already have safety teams. Neither is really about safety. Both are about capturing the institutional layers—data, deployment, governance—where AI value will actually accumulate. Smaller stories tell the story: Amazon backing formal verification for AI agents, Cornell's optical chips for edge robotics, Kuaishou's agentic coder trained on real codebases. The pattern suggests AI is maturing from theoretical capability to operational infrastructure, even as the geopolitical and institutional boundaries around it remain contested.

Read the full briefing →