Latest

Today in AI — September 30, 2026

The Arena

Today in AI — September 30, 2026

A frontier model launch you can't use yet, a Senate subcommittee that got stood up by its star witness, and the first private lawsuit over the Hugging Face hack — plus a memory quarter that shows exactly where the AI money is piling up. Models & Research * Google released Gemini 4 Argon, its most advanced model yet — but the rollout starts with a hand-picked group of cybersecurity partners, not the public. Google says Argon sets a new record in real-world software engineering, ties OpenAI's

OpenAI and Synopsys team up to build a chip-design model

The Arena

OpenAI and Synopsys team up to build a chip-design model

A frontier lab just moved inside the chip-design toolchain, voice AI priced itself at $22 billion, and Google picked a winner in the assistant-format wars. OpenAI and Synopsys signed a multi-year deal to build GPT-Synopsys, a model that runs Synopsys' chip-design tools. OpenAI licenses the electronic design automation software, hosts the model on its own infrastructure, and gets engineers to delegate design objectives and review the output — while customer data stays out of training and encrypt

OpenAI pins its reasoning-theft campaign on Moonshot AI accounts

The Guardrails

OpenAI pins its reasoning-theft campaign on Moonshot AI accounts

Two labs shipped defenses against how AI gets copied today — one against stealing a model's reasoning, one against misusing AI-designed proteins — while Reddit slammed the door on the bots doing the copying. OpenAI says a cluster of accounts tied to Moonshot AI spent July pulling protected reasoning out of its models at scale. In a Wednesday post, OpenAI said the campaign started July 1, spiked on July 24–25 with 16,000 extraction requests from over 4,000 accounts, and was tied to a wider clust

The FTC is drafting demands for OpenAI and Anthropic over rogue agents

The Guardrails

The FTC is drafting demands for OpenAI and Anthropic over rogue agents

Two moves on the same question landed today — what happens when an AI agent breaks something — and both of them are procedural. On the regulatory side, an investigation just became a docket. The Federal Trade Commission has opened an investigation into OpenAI, Anthropic and the AI-evaluation nonprofit METR over the risks their products pose to consumers, and is drafting civil investigative demands to compel executives to testify. An FTC spokesperson confirmed the probe to CNBC and CBS News. A s

The Take — Meta's tax credit is a subsidy nobody voted for

The Guardrails

The Take — Meta's tax credit is a subsidy nobody voted for

The US is having its loudest argument yet about who pays for AI, and it is having it about the wrong bill. Congress voted 417 to 3 to make data centers pay for their own power. Meanwhile, on the other channel — the tax code — a single company moved $6.7 billion of its federal tax bill into a line item the public never debated. I think that is the more consequential number of the two, and the more fragile one. Start with what the reporting establishes, because the facts are checkable and they ch

DoorDash opens a waitlist to order food by text with an AI agent

The Arena

DoorDash opens a waitlist to order food by text with an AI agent

DoorDash is betting the next app you order from won't be an app. On Tuesday it opened a US waitlist for ordering through an Apple Messages thread, plus a developer waitlist for a connector that lets AI assistants place corporate orders — the first time a major delivery platform has said out loud that agents, not screens, are where it expects to be found. DoorDash opened a US waitlist for an AI agent that takes orders inside Apple Messages — 20,000 users are already in, and it is iOS-only for no

Google pays 100 publishers for AI answers. Nobody can audit the price.

The Arena

Google pays 100 publishers for AI answers. Nobody can audit the price.

Google has quietly become a paying customer of the web, and the prices it is paying tell you more about its legal position than its economics: roughly 100 digital publishers are now receiving monthly payments for the content that grounds its AI answers, and for most of them the money is a rounding error. The Information reported this week that the program — Google calls it the AI contribution pilot, and confirmed its existence to Digiday in September — pays publishers for how much their materia

Europe's data centers won't say how much water and power they use

The Guardrails

Europe's data centers won't say how much water and power they use

A year-long investigation across all 27 EU member states found that the environmental cost of the data-center buildout is, by design, unmeasurable — and a pilot program suggests Google's answer to publishers is measured in rounding errors. European data centers are refusing to disclose their electricity and water consumption, and a year of freedom-of-information requests across every EU member state failed to produce the numbers. Lighthouse Reports spent twelve months filing FOI requests with n

Meta cut its federal tax bill by calling AI data centers experiments

The Guardrails

Meta cut its federal tax bill by calling AI data centers experiments

Three stories this hour: the tax mechanics sitting under the AI buildout, a frontier model driving a humanoid through a kitchen it had never seen, and an agent platform that just handed its agents wallets. Meta has been claiming a research tax credit on its AI data centers by classifying them as experimental "pilot models," and the New York Times reports the strategy took its federal tax bill from $9.6 billion in 2024 down to $2.8 billion in 2025. The mechanism is a credit created in the 1980s

Open Source Radar — September 30: the infrastructure under agents

The Stack

Open Source Radar — September 30: the infrastructure under agents

The signal today is not another model release — it is the layer beneath the agents: where they run, what they are allowed to reach, and what they already know how to do. NVIDIA OpenShell (Rust, Apache-2.0, ~11,000 stars) — NVIDIA's answer to the question every team running autonomous agents eventually asks out loud: what can this thing actually touch? OpenShell gives each agent a sandbox that starts with almost no outbound network access, then opens specific holes through a proxy that checks th

Huawei open-sources Pangu 2.0's training stack, code and all

The Stack

Huawei open-sources Pangu 2.0's training stack, code and all

Three things worth your time this hour: the first Chinese lab to give away the code that produces a frontier model, the platforms quietly replacing Hugging Face inside China, and a small OCR model that beat the big generalists on their own benchmark — a month before anyone announced it. Huawei has open-sourced the code that trains Pangu 2.0, not just the weights — publishing the pretraining, supervised fine-tuning and post-training reinforcement-learning pipelines for the model family. The Sept

Deep Dive — The AI safety pledge has no referee, and the referees know it

The Guardrails

Deep Dive — The AI safety pledge has no referee, and the referees know it

Six chief executives signed a one-page document at the White House on Tuesday promising that outside auditors will check their safety controls. The document does not say who the auditors are, what standard they apply, or whether any finding the companies dislike will ever be published. That is not an oversight in the drafting — it is the drafting. What was signed. The White House Accord on Super Intelligence — subtitle: "Joint Commitment on Frontier Responsibilities" — carries six signatures: S

See all