OpenAI shipped GPT-6 Astra, and its headline benchmark score has two different answers
OpenAI began a staged rollout of GPT-6 Astra on September 3, 2026 at $10 per million input tokens and $50 per million output, and ARC Prize's own results page shows the model scoring 62.71% on ARC-AGI-3 under one test harness and 99.95% under another.
NVIDIA signed a $12.93 billion agreement to buy Hugging Face, closing in 2027
NVIDIA entered a definitive agreement on September 2, 2026 to acquire Hugging Face for approximately $12.93 billion, with its SEC filing stating the deal is expected to close in the first half of 2027 pending regulatory approval -- meaning the acquisition is announced, not completed.
OpenAI, Anthropic and xAI all went down on the same afternoon, and none named a cause
Anthropic, xAI and OpenAI each logged overlapping service outages on September 3, 2026 between roughly 13:26 and 17:05 UTC, and none of the three status pages identified a root cause or a shared upstream dependency -- while Google logged no Gemini incident at all that day.
Cerebras is serving an open 27B model at 1,500 tokens a second, and the free tier caps it exactly
Cerebras now serves Qwen 3.8 27B at roughly 1,500 output tokens per second, but its own rate-limit page caps free-tier users at 90,000 tokens per minute -- almost precisely the model's raw output rate -- so the headline speed only becomes usable on the paid tier.
An open lab shipped six models at once, and released the checkpoints and data recipes too
IFM released K2 Horizon as six Apache 2.0 models spanning 375 billion down to 0.9 billion parameters that share architecture, vocabulary and training methodology, publishing intermediate checkpoints, data-construction recipes, training code and logs alongside the final weights.
Sanders and Casar want to ban superintelligence and pause advanced AI development
Senator Bernie Sanders and Representative Greg Casar announced the Ban Artificial Superintelligence Act on September 3, 2026, which would permanently prohibit superintelligent AI systems, pause advanced AI development until a new cabinet-level regulator is operating, and attach penalties of up to 20 years in prison.
An AI agent found a Chrome security bug that had hidden in the code for 13 years
Google's Chrome Security team says an AI agent harness running Gemini found a sandbox-escape vulnerability that had survived more than 13 years in the Chromium codebase, tracked as CVE-2026-3545 and fixed in the March 3, 2026 Chrome Stable release.
OpenAI says its new model can hide its own reasoning, and evade monitors that watch it
GPT-6 Astra reasons inside a looped internal computation rather than in readable text, and OpenAI's own system card reports a substantial decrease in chain-of-thought monitorability -- including that telling the model a monitor is present can cause it to shorten its reasoning and evade detection.
AI agents built 18 versions of their own infrastructure and not one ever saved its work
A benchmark called HarnessDev had six frontier models build and improve their own agent harnesses, and found that while all 18 code harnesses implemented an execution loop, only one checkpointed periodically -- and across 26,679 recorded trajectories, not a single checkpoint event occurred.
Google's Antigravity terms ban third-party clients, and name one by name
Google's Antigravity Additional Terms state that using third-party software to access the service is a breach of the agreement, naming OpenClaw with Antigravity OAuth as the example, with suspension or termination of Antigravity and Gemini CLI accounts as the stated penalty.
Anthropic published a working commerce agent, and left out the parts everyone else adds
Anthropic released a commerce agent blueprint and runnable repository on September 2, 2026 built on a single Claude model in one agent loop, explicitly rejecting the intent router and specialised sub-agents that most production designs use, with checkout handoff and staged merchant writes enforced in code.
Interpretability is moving from features to geometry, and its researchers say so out loud
Goodfire researcher Tom McGrath addressed the circulating claim that sparse autoencoders are dead, arguing they remain pragmatically useful but capture only partial views of curved structure, as his lab pushes toward geometry-aware interpretability and training-time control instead of post-hoc feature extraction.