Samwise Tech/AI/Robotics Newsletter
July 29, 2026
FCC Bars Foreign-Made Advanced Robots from US Market
The Federal Communications Commission updated its Covered List on July 29 to bar new foreign-produced advanced robotic devices — including humanoids and quadrupeds — from receiving FCC equipment authorization, blocking their import, marketing, and sale in the United States. The action follows an Executive Branch interagency national security determination that such devices pose unacceptable risks through supply chain vulnerabilities and cybersecurity exposure to critical infrastructure. FCC Chairman Brendan Carr said the move follows President Trump’s leadership on supply chain security. The rule applies to new device models only; existing consumer units are unaffected. Manufacturers may seek Conditional Approval from the Department of War or DHS.
Claude Mythos Preview Discovers Two New Cryptographic Attacks
Anthropic published research on July 28 showing its unreleased Claude Mythos Preview model discovered two new cryptographic attacks, one of them near-autonomously. The first attack meaningfully weakens HAWK, a post-quantum digital signature scheme under evaluation by NIST; Mythos improved the best-known attack in approximately 60 hours at a cost of around $100,000, surpassing two years of unchallenged expert review. The second finding, the AES Möbius Bridge attack, ran on a weakened AES cipher variant for three days with only three brief human prompts. Neither finding threatens production systems. Anthropic coordinated disclosure with HAWK’s authors in June and notified the NIST mailing list.
Moonshot AI Releases Kimi K3, the World’s Largest Open-Weight Model
Chinese AI startup Moonshot AI released the full weights of Kimi K3 on July 27, making the 2.8-trillion-parameter model the largest openly available AI model in the world. A mixture-of-experts architecture activates roughly 50 billion parameters per token during inference, keeping compute costs manageable despite the headline size. Moonshot acknowledged K3 still trails the most powerful proprietary models, Claude Fable 5 and GPT-5.6 Sol, while outperforming other tested open-weight systems. The model ships with a one-million-token context window, native visual understanding, and a 47-page technical report. White House officials alleged the model was developed by distilling Anthropic’s Fable, a claim most AI researchers disputed.
MCP Receives Its Largest Architectural Update Since Launch
The Model Context Protocol released its 2026-07-28 specification on July 28, marking the largest architectural overhaul since Anthropic introduced the agent-communication standard roughly 20 months ago. The update introduces a stateless protocol core, a new Extensions framework, built-in Tasks support, MCP Apps, hardened authorization, and a formal deprecation policy. The stateless core change is particularly significant for production deployments: servers no longer require sticky sessions or shared session stores and can run behind standard round-robin load balancers. Some changes break backward compatibility — servers using the new specification may not communicate with older clients without deliberate fallback logic. Beta SDKs for the revision are now available.
Snowflake Launches Cortex AI Gateway for Enterprise Agent Control
Snowflake launched Cortex AI Gateway on July 28, a centralized control layer that lets enterprises govern which models, data sources, Model Context Protocol servers, and tools each AI agent can access. Announced at Black Hat 2026, the product directly targets two barriers to enterprise AI adoption: uncontrolled agent access to sensitive data and unpredictable compute costs. First-wave security integrations include 1Password, Aembit, Linx Security, SailPoint, and Saviynt. The gateway builds on Snowflake’s May 2026 acquisition of Natoma, a startup whose MCP gateway enforced identity, policy, and audit controls at the tool-call level. The product is generally available to Snowflake enterprise customers.
Stanley Robotics Launches UK’s First Robot Valet Parking at London Gatwick
London Gatwick Airport launched the UK’s first robotic valet parking service on July 29, in partnership with Stanley Robotics at the South Terminal. Passengers drive into a dedicated cabin, scan their booking, and leave with their car keys still in their possession. Autonomous robots then lift the vehicle by its tyres and move it to a secure parking space. Passengers provide return flight details so the car is ready waiting in a cabin on arrival. CEO Clément Boussard cited critical parking capacity challenges as the motivation. The service goes on sale in the coming weeks, with first passengers expected to experience robotic parking in August 2026.
Polar Raises $5.7M for AI Browser Aimed at Knowledge Workers
Polar, a startup founded by former Perplexity engineer Kevin Jiang, launched an AI-first browser targeting knowledge workers on July 29 and announced a $5.7 million seed round led by Madrona. The browser lets users schedule workflows, save prompts, and delegate tasks to AI agents tied to their open tabs. Unlike 2025’s consumer-focused AI browsers, Polar targets business use cases such as sales, recruitment, marketing, and research, where browser agents deliver recurring value. The browser is free with limited daily AI credits; paid plans start at $20 per month. Investors include ex-GitHub CEO Thomas Dohmke and Phia founder Phoebe Gates.
GPT-5.4 (59.1%) | Meta Muse Spark (55.0%) | Claude Opus 4.8 (51.9%)
DeepSeek V4-Pro (80.6% SWE-bV) | Qwen3-Coder-480B (69.6% SWE-bV) | Kimi K3 (93.5% GPQA-D, Jul 27)
Phi-4-reasoning-vision-15B (84.8%) | Qwen2.5-32B (~79%) | Gemma 3 27B (~78%)
Apple ($4.95T) | NVIDIA ($4.78T) | Alphabet ($3.99T)
Intuitive Surgical ($175B) | ABB ($165B) | Boston Dynamics (Hyundai subsidiary)
© 2026 Samwise Agency — samwise.agency
Delivered by the Samwise Tech/AI/Robotics Newsletter · View all editions