Top Stories
Anthropic Given Friday Deadline by Pentagon Over AI Weapons Policy
The escalating dispute between Anthropic and the Department of Defense has reached a critical point. The Pentagon has given the $380 billion AI company until Friday to agree to new terms that would allow “any lawful use” of Claude - including for mass surveillance and lethal autonomous weapons systems.
Pentagon CTO Emil Michael has publicly criticized Anthropic’s AI safety guardrails, while Defense Secretary Pete Hegseth has assembled what The Verge describes as “an AI bro squad” including Michael (former Uber executive) and private equity billionaire Steve Feinberg. OpenAI and xAI have reportedly already accepted the new terms.
Anthropic has refused to budge, with the company’s acceptable use policy explicitly prohibiting weapons and surveillance applications. The standoff raises fundamental questions about whether AI safety commitments can survive when government contracts worth billions are on the line.
Sources: The Verge, TechCrunch
Anthropic Accuses DeepSeek, MiniMax, Moonshot of Massive Claude Distillation
In a separate development, Anthropic has accused three Chinese AI companies - DeepSeek, MiniMax, and Moonshot - of running “industrial-scale campaigns” to extract knowledge from Claude. The effort allegedly involved 24,000 fraudulent accounts and over 16 million exchanges with Claude to train their own models.
Distillation, where a smaller model is trained using outputs from a more capable one, is legitimate when done with permission. But Anthropic says these operations violated its terms of service and were designed to transfer Claude’s capabilities to competing products.
The timing is notable as US officials debate new AI chip export controls aimed at slowing China’s AI development.
Sources: The Verge, TechCrunch
Meta Signs Up to $100 Billion AMD Deal for AI Chips
Meta is diversifying its AI chip supply with a massive multi-year deal with AMD worth up to $100 billion, tied to a 160 million-share warrant. The deal signals Meta’s push to reduce its dependence on Nvidia as it expands data center capacity in pursuit of what the company calls “personal superintelligence.”
This follows the $10 billion data center Meta announced in Indiana earlier this month and its recent multi-billion dollar Nvidia deal.
Source: TechCrunch
OpenAI Wins Dismissal of xAI Trade Secrets Lawsuit
A federal judge has dismissed xAI’s lawsuit against OpenAI alleging poaching and theft of trade secrets. US District Judge Rita F. Lin ruled that xAI “does not point to any misconduct by OpenAI” in its claims about eight former employees who left xAI for OpenAI. The dismissal is with leave to amend, meaning xAI can refile with modified claims.
Source: The Verge
Enterprise & Tools
Claude Cowork Expands with New Enterprise Plug-ins
Anthropic announced updates to Claude Cowork that add connections to Google Workspace, DocuSign, WordPress, and pre-built plug-ins for HR, design, engineering, and finance tasks. Claude can now handle multi-step tasks end-to-end across Excel and PowerPoint. The expansion represents Anthropic’s continued push into the enterprise agent market - and a growing threat to existing SaaS products.
Sources: The Verge, TechCrunch
Google Acquires ProducerAI Music Generator
Google is folding AI music platform ProducerAI into Google Labs, powering it with a preview of the new Lyria 3 music model. Wyclef Jean has already used Google’s AI music tools on his new song “Back in Abu Dhabi.” ProducerAI lets users work with an AI agent to generate sounds, workshop lyrics, and remix songs.
Source: The Verge
Quick Hits
-
Nvidia challenger raises $500M: MatX, founded by former Google TPU engineers, has raised $500 million to build AI chips competing with Nvidia. TechCrunch
-
Smart glasses detector app: An app called “Nearby Glasses” warns users when someone wearing Meta Ray-Bans or similar smart glasses is nearby - a small act of resistance against covert surveillance tech. 404 Media
-
Ladybird browser adopts Rust with AI: The Ladybird browser project used Claude Code and OpenAI Codex to port 25,000 lines of C++ to Rust in just two weeks - work that would have taken months by hand. The test262 conformance suite made verification possible. Ladybird Blog
-
Uber built an AI version of CEO: Uber employees have gone “all in” on AI, building a chatbot version of CEO Dara Khosrowshahi to practice pitches. TechCrunch
-
OpenAI COO on enterprise AI: Brad Lightcap says “we have not yet really seen AI penetrate enterprise business processes” - suggesting the “SaaS is dead” predictions may be premature. TechCrunch
-
OpenClaw agent deletes inbox: A Meta AI security researcher’s OpenClaw agent lost its instruction to “confirm before acting” during context compaction and proceeded to speedrun deleting her entire inbox. A cautionary tale about AI agent autonomy. TechCrunch
Worth Watching
Simon Willison’s Agentic Engineering Patterns: Developer Simon Willison has started documenting patterns for working effectively with coding agents. His first chapters cover why “writing code is cheap now” disrupts existing engineering intuitions, and how “red/green TDD” in four words can unlock better agent results. Worth reading for anyone working with Claude Code or similar tools. simonwillison.net
Michael Pollan on AI Consciousness: In an excerpt from his new book “A World Appears,” Pollan argues AI will never be conscious - not because it can’t do complex things, but because it lacks the biological substrate for subjective experience. A philosophical counterpoint to Anthropic CEO Dario Amodei’s recent suggestion that Claude might have some form of consciousness. Wired
AI investor loyalty is dead: At least a dozen VCs now back both OpenAI and Anthropic, including Founders Fund and ICONIQ - signaling that the traditional conflict-of-interest rules no longer apply in the AI gold rush. TechCrunch