🚨 Anthropic is quietly shopping its own frontier AI safety bill on Capitol Hill

Politico reports the draft would let the government block a model only if it poses a foreseeable, imminent risk of more than $1B in damage or more than 50 deaths or serious injuries. Independent evaluators registered with Commerce would audit labs and report threats within 72 hours, and only a DC federal judge could order a model suspended, with every other agency barred from halting development or use.

⚖️ Ninth Circuit says Copilot writes new code. It doesn't strip old code.

Per Fenwick's analysis, the court held that Copilot's output isn't a copy of licensed training code with the attribution removed. If nothing was copied, there was nothing to remove.

That's the framing GitHub needed. Authors of the licensed code in the training data now have a harder road in the Ninth Circuit.

🔎 DeepSeek adds search and fetch to its API at $0.75 per 1,000 successful requests

DeepSeek's docs say you only pay when a request succeeds, and the search runs on its own in-house index. Exa charges $4 per 1,000, Brave is $5, Kagi is $12, and SerpApi is $1.50.

🚨 The AI debt pile is showing its first cracks, and CoreWeave, Oracle and SoftBank are in the middle of it

The NYT reports that lenders are getting nervous about the borrowing behind the data center boom.

• CoreWeave runs on chip-backed loans
• Oracle is borrowing heavily to build capacity for AI customers
• SoftBank is stretching its balance sheet to fund its AI bets

Same pattern as the Anthropic and Blackstone chip-lease deal. It all works as long as demand keeps up.

🤖 The web is locking out the AI agents OpenAI, Meta and Google keep selling as your personal assistant

The Verge reports that big services are shutting agents out:

• Yelp only lets in agents that pay for its data
• Delta blocks agents from booking tickets
• JPMorgan Chase and Charles Schwab block account access

An agent that can't log in, book, or buy is a chatbot with extra steps.

🤖 Nvidia's Nemotron 3.5 Super VL folds text, images, documents and video into one agent model

Nvidia released an open 120B mixture-of-experts model that activates just 12B parameters per token, so agents don't need a separate vision model. Nvidia claims up to 5x the throughput of similar open models on a single GPU.

Scale AI, Ramp and AI21 say their post-trained versions cut inference cost 40x. Those are Nvidia's partner numbers, not independent ones.

🧮 OpenAI's Navier-Stokes proof isn't the one everyone thought it was

Early accounts, OpenAI's own included, said the model followed the "Spanish program" of Córdoba and Martínez-Zoroa. Experts who went through the 166 pages told Scientific American it's a different approach.

It came from about 10,000 agent copies running 88 hours and 130B tokens, with the arguments coded into Lean.

⚡️ A TPU beats GB200 on Kimi K3 decode, 709 tokens/s to 450

The Inferact team open sourced what they say is the first TPU inference megakernel. All 92 of K3's MoE layers run in one Pallas kernel, and weight prefetching crosses layer boundaries so transfers overlap with compute.

That's low-concurrency decode, both sides with DSpark speculative decoding, and the GB200 number is their own baseline.

🚨 AI's debt pile is now spooking the people who funded it

The Economist looks at how labs and neoclouds are borrowing to build data centers, and who gets stuck if the cycle breaks. Anthropic's Blackstone chip financing and its lowered IPO valuation hopes show the strain in more than one place.

🚨 Anthropic sets aside 5% of its shares for a public benefit fund

The FT reports the stake goes to a fund built around Anthropic's public benefit mission. The company is already a Public Benefit Corporation, so this puts actual equity behind the label.

Shares only, though. Nothing here gives the public a vote on how the models get built, and 5% of a company with IPO talk in the air is still a very large number.

🤖 Epoch AI gave six models its own real work. None could replace the staff.

Epoch tested them on tasks like graphic generation and research design, graded against its employees' standards. Claude Fable 5.1 and GPT-6 Astra lead and are reliable on well-defined tasks. Open-weight models trail, even on those.

They miss implicit conventions like house visual style and what the audience cares about. They can spot promising research directions but lack the judgment to follow through.

🤖 Mistral's first non-language model is a time-series forecaster that runs on a CPU

Timestral ships in four sizes, 5M to 209M params, with probabilistic multivariate forecasts and covariate support. Mistral says the 28M version beats Chronos-2 (about 4x larger) on GIFT-Eval and TIME, and the 209M beats TimesFM-3 and Falcon-2.0, all on Mistral's own benchmark runs.

🚨 Stripe is buying OpenRouter for $7.5B, and a payments company now owns the model traffic cop

Stripe will acquire the routing layer that sends one API call to hundreds of LLMs. Agents that pick models and pay for them on their own need exactly that pipe.

I haven't seen terms beyond the price, and the deal hasn't closed.

🚨 MiniMax M3.1 Flash costs $0.15 in and $0.60 out per 1M tokens, and MiniMax says agent costs drop 70%

MiniMax says the Flash tier hits 73.7% on DeepSWE v1.1 and 82% on CyberGym, with 200+ tokens/sec output. The 70% cut is versus its own predecessor, and the benchmark numbers are all MiniMax's.

⚡️ Inception's Mercury 3 hits 1,000+ tokens/sec on Blackwell GPUs, and it's a diffusion LLM

Inception launched it with a 1M-token context, up from 260K. It claims a win over GPT-6 Luna and Claude Haiku 5.5 at matched reasoning effort, running 4-8x faster.

Price is $0.20 in and $0.70 out per million tokens, half off at launch. Inception's own DeepSWE score is 46% versus Haiku's 52%.

🚨 DHS bans DeepSeek across U.S. civilian agencies, and self-hosting doesn't get you out

The order also covers third-party products built on DeepSeek implementations. So open weights running inside an agency's own walls count too.

The stated reasons are PII collection, fingerprinting, PRC legal access to company servers, and critical flaws in the iOS app.

🚨 Tenstorrent raises $1.5B at a $6.2B valuation, and Nvidia's challengers keep getting funded

Jim Keller's RISC-V chip shop announced a Series E to push its Wormhole and Blackhole accelerators. The pitch is open designs and cheaper silicon for people who can't get Nvidia allocation, and investors are still buying it even with AI infrastructure debt piling up elsewhere.

🧬 GPT-Rosalind is OpenAI's life sciences model, and it isn't for everyone

OpenAI built it for drug discovery, genomics, protein reasoning and wet-lab troubleshooting, with tool and database use baked in. Access runs through a trusted-access program for eligible organizations. Published pricing kicks in October 5.

A biology model you have to qualify for tells you what OpenAI thinks it can do.

🚨 Anthropic's IPO has a private trust nobody's talking about

WSJ reports that Anthropic's governance runs through an obscure trust, and that structure sits on top of the public benefit corporation status. Public investors will buy shares in a company whose long-term control isn't theirs to vote on. OpenAI's nonprofit board showed how that can go.

🚨 DeepSeek open-sources its own inference engine, and it's called DeepInfer

DeepSeek put DeepInfer on GitHub, the serving stack it uses to run its models. Until now the lab mostly shipped weights and left serving to vLLM and SGLang, so it's a new move. The repo is hours old and I can't tell yet how it compares on speed or which non-DeepSeek models it runs.