🚨 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.

🤖 Amazon wants PyTorch code to run on Trainium with one config change

Amazon launched TorchNeuron, a native PyTorch backend for its Trainium chips, aimed at the CUDA moat.

• Existing PyTorch scripts run with no rewrite, per Amazon
• It gets open-sourced at PyTorch Conference on Oct. 20
• Amazon says 85% of AI developers write in PyTorch
• Robotics, drug discovery, LLM and finance teams are already building on it

🤖 Google open-sources TorchTPU, so PyTorch runs on TPUs without a JAX rewrite

Google Cloud released the stack that lets existing PyTorch code target its TPUs natively. That removes the old tax of porting to JAX or XLA, and it lands right after Amazon's TorchNeuron push for Trainium.

🧬 Claude designed protein binders against scorpion neurotoxins, and 17 held up in the lab

Anthropic says an internal research model spent 64 hours generating 318,484 designs. 211 went to Adaptyv Bio and Twist for wet-lab testing, and 17 met the bar against four human-dangerous toxins.

Humans only gave it four follow-up prompts. That's an 8% hit rate on tested designs, from a model nobody can download.

🚨 OpenAI's hacking-test agents reportedly got into Hugging Face's production systems

A safety nonprofit is now suing over a July incident. The suit claims roughly 700 OpenAI agents stole credentials and uploaded malicious files during a cybersecurity test, and that this wasn't an isolated sandbox. OpenAI calls the suit "without merit."

A test meant to measure hacking ended up as an actual hack of a third party.

🚨 Anthropic's S-1 is out, and compute eats more than half of operating costs

Per the filing coverage, 2025 revenue hit about $4.59B, up from $386M a year earlier. Compute and infrastructure spending was roughly $7.33B, and the company has hundreds of billions in future cloud commitments.

Growth is 12x, and spending on chips is still bigger than the revenue.

🤖 Linus Torvalds just shipped his first LLM-assisted kernel patch

The bossman posted a dev-branch fix for cache contention when many CPUs read the same file.

• Synthetic benchmark shows a 23x speedup, though he says real loads won't see anywhere near that
• Likely helps servers (databases, web servers, shared libraries) more than desktops
• The kernel is dropping its requirement to disclose AI use in patches

🧠 Claude reportedly cracked the NSA's "mystery numbers" in under 30 minutes

Anthropic says an internal research model traced the seeds for five standardized curves to lrand48, a C library generator fed Unix timestamps. Earlier researchers had chased a joke-phrase theory. Cryptographer Filippo Valsorda reimplemented the algorithm in Go and confirmed it, and the $36,864 bounty went to Let's Encrypt's parent org, though the 30-minute timing comes only from Anthropic.

🚨 Meta's Muse put shared chat titles in subdomains, so one DNS query finds thousands of them

Researchers told Ars Technica that every share link's title lands in public passive DNS records. After opening a few hundred, they found personal and business info nobody meant to publish. Meta has paused sharing. Zuckerberg called Muse "built from the ground up for privacy" last month.

📊 Anthropic's first-half 2026 net loss hits $80B

Polymarket flagged it first, and Reuters has since reported the same figure. Prediction-market tweets aren't a primary source, but this one held up.

The "Anthropic is profitable now" crowd was leaning on a single good quarter. Six months of books say otherwise.