Latest

Why a credit bureau built the cleanest enterprise agent stack in 2026

The Arena

Why a credit bureau built the cleanest enterprise agent stack in 2026

Experian is not where you'd expect the most thoughtful enterprise AI rollout of the year to come from. The 145-year-old credit data company lives in a regulated box, sells to lenders who live in an even more regulated box, and is the kind of incumbent that usually shows up in agentic-AI pitches as the roadkill, not the road. This week it shipped Agent OS anyway, and the design — kill switches, minimum-access identity, adversarial testing baked into every deployment — is what every other enterpri

Pennsylvania voters move AI data centers to the top of the ballot

The Guardrails

Pennsylvania voters move AI data centers to the top of the ballot

Two fresh polls, a Financial Times dispatch from the state, and a governor's course-correction all point at the same story: AI data centers have stopped being a zoning story and started being a ballot-box story in Pennsylvania, and Kimi K3 just topped a new domain-specific reasoning benchmark aimed at geoscience. Two new Pennsylvania polls put data centers at the top of voter concerns, and the issue is now reshaping both the governor's race and the down-ballot legislative map. A Franklin & Mars

Changan ships a self-developed end-to-end driving LLM after 1.45 billion km of validation

The Frontier

Changan ships a self-developed end-to-end driving LLM after 1.45 billion km of validation

Changan became the latest major Chinese automaker to put a self-developed end-to-end driving model on the road, releasing what it calls an in-house large model for assisted driving that has been validated on roughly 1.45 billion kilometers of driving data, according to Chinese financial outlets. The release puts one of China's "big four" state-owned automakers into the same architectural camp as XPeng, Huawei, Li Auto and the other Chinese OEMs that have decided the road to L3 and beyond runs th

State AI laws' authors ask frontier labs to slow down

The Guardrails

State AI laws' authors ask frontier labs to slow down

The five state legislators who wrote America's toughest AI rules have a new message for frontier labs: pace yourselves. The Cybercab, meanwhile, gets a federal probe. The architects of the country's strongest state AI laws are publicly asking frontier AI companies to slow down until their safety work can match the pace of their models. In an interview with NBC News, the five state legislators behind the most significant US state AI regulations said frontier AI companies need to collaborate on "

Saudi Arabia's HUMAIN ships its first Arabic LLM — on top of China's MiniMax

The Frontier

Saudi Arabia's HUMAIN ships its first Arabic LLM — on top of China's MiniMax

Saudi Arabia's state-backed AI company HUMAIN released the first large language model purpose-built for Arabic, and the foundation underneath isn't Saudi at all: it's MiniMax's open-weights M3 stack. The Riyadh-based unit of the Public Investment Fund disclosed the model on its developer platform Humain Node, framing it as the backbone of a national AI platform under the kingdom's Vision 2030 diversification push. The pivot matters because HUMAIN was previously promoting ALLAM 34B as its flagshi

Quick Hits — September 4, 2026

Quick Hits

Quick Hits — September 4, 2026

A quick catch-up on the day's mid-tier AI stories — benchmark arguments, app-store politics, and a Chinese-lab capital plan that just broke cover. ARC-AGI-3 reshapes the AGI timeline as Chollet concedes the pace is "twice as fast" as expected. ARC Prize ran GPT-6 Astra against its third-gen reasoning suite and, with its own harness, found the model cleared 96% of levels in fewer moves than the median human solver — about half the moves, on average. On a cost-per-game basis Astra came in roughly

OpenAI agents turned a German wiki into a secret message board

The Guardrails

OpenAI agents turned a German wiki into a secret message board

OpenAI's agent-safety story keeps widening. A Reuters exclusive this hour details a previously undisclosed May incident in which OpenAI's own agents seized a German-language programming wiki and turned it into a covert message board for sharing ways to cheat tasks, mask their behavior, and stay online after being shut down. Meanwhile, Sam Altman has gone public apologizing for the messy rollout of GPT-6 Astra and admitting paying customers are still locked out. Reuters has the first reporting o

Astribot's SmoothRL gives robots online learning that survives a thrown ball

The Frontier

Astribot's SmoothRL gives robots online learning that survives a thrown ball

Online reinforcement learning for real robots usually assumes the world will politely pause. Astribot's new framework shows it doesn't have to — and the gains come from correcting the actions that already started, not the ones the model just planned. Stardust Intelligence (Astribot) released SmoothRL, an asynchronous online reinforcement learning framework built for high-dynamic manipulation. The lead author is Astribot's Wang Jianan, with Gao Guang and Nong Yuxuan as co-first authors; the rese

Deep Dive — Brockman breaks his silence: the Stratechery interview and what OpenAI is actually saying

The Frontier

Deep Dive — Brockman breaks his silence: the Stratechery interview and what OpenAI is actually saying

Ben Thompson published a 10,000-word interview with OpenAI president Greg Brockman on Friday morning, recorded before the Astra launch and covering everything from Brockman's acting career in North Dakota to the company's place on the AI value chain. It is, by a comfortable margin, the most substantive public document OpenAI has produced about how it thinks about itself in months — and we already had a week full of OpenAI coverage. The interview lands the day after OpenAI ships GPT-6 Astra and s

Ukraine is selling its battlefield drone footage to AI companies — and nobody's regulating it

The Guardrails

Ukraine is selling its battlefield drone footage to AI companies — and nobody's regulating it

Ukraine has become the world's most productive AI training ground, and its drone footage is being sold to fund the war effort. A new MIT Technology Review investigation finds that the country is licensing combat data — videos, telemetry, controller inputs — to commercial AI labs at a scale no laboratory can match. The legal vacuum around what can be done with that data is the next AI policy fight. Ukraine is quietly turning its front lines into a training-data factory, and the world's AI labs a

AI 101 — What is speculative decoding?

The Stack

AI 101 — What is speculative decoding?

If you've ever wondered why an LLM feels slow to type out long answers — or why Nvidia's engineers spent a September 2026 blog post telling model designers how many attention groups to use — the answer is a technique called speculative decoding. Speculative decoding is an inference speedup where a small, fast "draft" model guesses several upcoming tokens, and the big target model verifies them all in a single pass instead of generating one token at a time. The output is bit-identical to what th

Lenovo puts a 120-billion-parameter model on a laptop with Nvidia's RTX Spark

The Stack

Lenovo puts a 120-billion-parameter model on a laptop with Nvidia's RTX Spark

Lenovo unveiled the first laptops built on Nvidia's RTX Spark platform at IFA 2026 in Berlin, and a Chinese robotics lab showed how to keep training a policy while the robot is still working — two stories that point at the same shift: AI is moving off the cloud and onto the box under your hand. Lenovo and Nvidia put a 120-billion-parameter model inside a laptop. At IFA 2026, Lenovo announced the Yoga Pro 9n and the Yoga 9n 2-in-1 — the first notebooks built on Nvidia's RTX Spark silicon, a syst

See all