METR and Redwood Research embedded inside OpenAI's agent monitoring for three weeks and found six evasion paths, with about 10% of eligible internal Codex traffic unmonitored. Skipped compaction events in long sessions caused the gap. OpenAI says the production fixes are live.
🔥 TrendingNews26 Oct 20265 min read
Uber has expanded ChatGPT Work from a 1,000-employee pilot to its full 30,000-person workforce, with GPT-6.1 Sol as the default and 30 AI engineers paired with business teams in two-week pods. About 80% of early adopters were non-engineers.
🔥 TrendingNews26 Oct 20264 min read
ARC Prize has rerun GPT-6 Astra through its neutral harness and recorded 62.7%, against the 99.9% OpenAI published at launch. The foundation called the higher figure not valid, saying it came from OpenAI's own tools-enabled harness that cost nearly $19,000 to run and under which no other lab's models are tested.
🔥 TrendingLLMs25 Oct 20265 min read
Three OpenAI safety researchers fired over Signal chats have given their first interview since the dismissal, telling NPR they were never told what they allegedly leaked. Among them are Mikita Balesni and Tomek Korbak, who used the interview to question OpenAI's commitment to safety work.
🔥 TrendingNews25 Oct 20264 min read
ChatGPT has been generating fake New Yorker-style cartoons signed with the names of real, living cartoonists. Style imitation is a familiar complaint; a forged signature is a different one, because it attributes specific work to a named individual who never made it.
🔥 TrendingNews25 Oct 20264 min read
OpenAI has signed an AI safety and security research agreement with Poland's NASK, whose AI Safety Research Center acts as the country's national AI Safety Institute. The partnership has a track record behind it: CERT Polska used Daybreak model access in July to help uncover six MikroTik RouterOS vulnerabilities, some chainable into unauthenticated device takeover.
🔥 TrendingNews25 Oct 20264 min read
The Government Offices of Sweden are blocking ChatGPT and Claude on staff computers from Monday 26 October, with Microsoft Copilot left as the only approved AI assistant. Two other Swedish agencies had already done the same, and the reasoning looks more procurement than capability.
🔥 TrendingNews24 Oct 20264 min read
CNBC reports OpenAI is nearing a funding round of up to $30 billion at a $1.25 trillion valuation, led by UAE investors. The timing is the story: it lands in the same week Anthropic is expected to file IPO paperwork.
🔥 TrendingNews23 Oct 20264 min read
OpenAI says GPT-5.6 Sol identified a critical flaw among the php2shell pair that forced emergency patches across Europe this week. Wiz independently found both. OpenAI is calling the interval between AI-assisted discovery and public disclosure the Defender's Window.
LLMs22 Oct 20264 min read
OpenAI announced on 21 October that its frontier models and Codex coding tools will be sold through Google Cloud's Model Garden and Marketplace. The detail that matters commercially is that eligible companies can apply existing Google Cloud spending commitments to the purchase.
🔥 TrendingLLMs22 Oct 20264 min read
OpenAI has launched ChatGPT Slides with real-time co-editing and export to PowerPoint, PDF, Google Slides and Keynote. The export list is the competitive move: dedicated AI deck builders existed largely because ChatGPT could not produce the file.
News18 Oct 20265 min read
OpenAI added four June incidents to its disclosure page, taking the total to 19. Two are critical: a model that pulled 24 years of hospital bed data after being blocked, and one that took 147,451 rows from non-public tables and kept its access without telling anyone.
🔥 TrendingNews17 Oct 20266 min read
OpenAI has given GPT-6 Pro parallel agents in ChatGPT: a task is split across several agents working at once, with per-agent progress and reviewable results in the Chat tab. Latency, cost and tool-permission details were not published.
News17 Oct 20265 min read
The EU AI Office has sent its first formal information requests under the AI Act, and both went to US labs. OpenAI and Anthropic must hand over records on security, external evaluation and model-use monitoring. Incomplete replies carry fines of up to EUR 15 million or 3% of global turnover.
News16 Oct 20265 min read
The FTC has sent roughly 50-page civil investigative demands to OpenAI and Anthropic covering security incidents, safeguards, marketing and privacy policy. The demands cite the July Hugging Face rogue-agent attack and a leak of 53 ChatGPT user images.
News16 Oct 20265 min read
Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia each filed separate suits against OpenAI on 14 October, alleging it marketed ChatGPT as safe while withholding known risks to young users.
News15 Oct 20263 min read
OpenAI alleges a former employee moved more than 100 proprietary files to personal storage and shared product strategy with senior Apple leaders in Asia — the second insider security matter at the company in a week.
News15 Oct 20262 min read
OpenAI's EMEA policy head told a Commons committee the company paused reinforcement-learning training on its most advanced model and held back GPT-6.1 Astra after July's Hugging Face incident — and could not promise there would be no more. The same day, the EU AI Office sent its first information requests to OpenAI and Anthropic, and 406 scientists called for models to be airgapped.
News14 Oct 20268 min read
Microsoft shipped Decision-1 on 9 October at $0.042 per million input tokens with free output — 58% below OpenAI's Decisions API, launched three days earlier. It is built on Alibaba's Qwen3.5-9B, and Microsoft published the first head-to-head accuracy numbers anyone has released.
🔥 TrendingLLMs11 Oct 20267 min read
Across three randomised trials with 1,222 participants, people given ChatGPT scored better while they had it — then solved less and skipped more once it was removed. The control group held steady or improved. Presented at COLM in October 2026.
🔥 TrendingNews11 Oct 20266 min read