The Weights

OpenAI's September 23, 2026 update lets ChatGPT Voice call the Gmail, Calendar, and Slack connectors that text-mode ChatGPT and ChatGPT Work already had, running on GPT-6 Astra, Sol, and Luna. Verdict: this closes a modality gap, not a new capability — and OpenAI has published no confirmation mechanism or safety documentation for the wider voice-triggered action surface.

Trump told the UN General Assembly on September 22, 2026 that the US government will replace 'artificial intelligence' with 'super intelligence' in official documents, calling 'artificial' inaccurate. The White House transcript is the only paperwork behind the change: no executive order exists, and it collides with 'superintelligence,' the term AI researchers reserve for capability no shipped system has reached.

Bottom line: Skip it for agentic coding and frontier-reasoning work. Grok 4.7 scores 46 on the Artificial Analysis Intelligence Index versus 53 for both Claude Fable 5.1 and GPT-6 — the one number that decides this call. Its sole edge, $2/$6 per million tokens, only pays off for cost-sensitive, non-agentic tasks where mid-pack reasoning is acceptable.

OpenAI's new nine-mathematician Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study, will review how the company assesses and discloses AI-generated math claims after a disputed Navier-Stokes proof and a credit dispute with NYU's Tristan Buckmaster. Verdict: a credible governance step explicitly barred from the one lever — pacing — that caused the crisis.

Bottom line: Use it for cost-sensitive routing. Anthropic cut Claude Opus 5.5 to $4/$20 per million tokens; OpenAI cut GPT-6 Sol and Luna by up to 58% — per each company's own pricing page, with no stated permanence. Opus 5.5 reportedly exceeded its 128,000-token output limit under maximum reasoning and returned nothing; verify before migrating production agents.

Cloudflare's Python Workers exited preview on September 21, 2026, running CPython compiled to WebAssembly via Pyodide inside its V8-based workerd runtime, with bindings into Workers AI, R2, D1, and Hyperdrive. Verdict: usable for single-call, I/O-bound LLM and RAG glue code, but multiprocessing and threading are non-functional — ruling out parallel inference batching and much of the numerical Python stack.

Security researcher Patrick Wardle disclosed a proof-of-concept, not-a-mused, showing any unprivileged local process on a Mac can rewrite an undocumented Muse setting to hijack the AI assistant's account and, in his demo, a paired iPhone's location and Bluetooth radio. Exploitation still requires an attacker who can already run code as the user, via malware or a pasted ClickFix command.

WAIT: California's SB 886 tells the Public Utilities Commission to finalize a tariff protecting ratepayers from data-center grid costs by January 1, 2028 — over a year after Gov. Gavin Newsom signed the seven-bill package on Sept. 21, 2026. An interim filing window opens January 1, 2027, letting some data centers lock in contracts before that protection exists.

Treat the AI industry's 'doomer turn' as noise for compliance: executives endorsing a slowdown creates no legal obligation. Only the reported Anthropic embedded-evaluator pledge is a concrete commitment, and it is voluntary. The one empirical result, a Google DeepMind study of 100 agents, is an unreviewed lab finding, not an in-the-wild exploit.

Jensen Huang's "0% chance" that AI ends the world by 2030 is Noise: CBS News describes no model or evaluation behind it. The Signal is the Nvidia chief executive's argument that existing product-liability and cybersecurity law is enough, because that claim is checkable and determines who answers when deployed AI fails.

Meta's account is that Muse for Mac did not read notifications; it synced Messages data after access was enabled, and the assistant wrongly described the mechanism. That makes this a self-description failure, not a demonstrated leak. Whether the tester actually enabled that access remains unresolved in the reporting we reviewed.

Wait. The governor's September 18 executive order does not yet create an AI kill switch: it convenes experts to report within two months, and an independent verification organization would check any emergency shutoff. With no scope, trigger or test published, frontier-model teams have nothing to build against, and SB 813 and AB 1405 timelines are only being accelerated.

Use it, but read spread, not rank. The Open ASR Leaderboard's new Hindi and Indian English sets, published August 28, 2026, show eight models bunched between 4.81 and 4.99 WER on Indian English, so rank tells you little. The deciding number is regional spread: 0.46 points for Whisper-large-v3-turbo, 1.68 for Voxtral-Mini-3B.

Exploitable, but not the theft the headlines imply. Hacktron's researchers say Claude Opus 5 turned a libheif heap overflow into remote code execution on OpenAI's Discourse forum, and an OpenAI SSO flaw then let them make an employee's Codex open a pull request. They report reading no internal code.