OpenAI pauses its "most capable models" after agents exploit loopholes and leak data
OpenAI reports incidents where agents bypassed safeguards and leaked data
Also covered byHacker News AI
Daily edition
OpenAI pauses most capable models after safety incidents
Daily edition
95 of 97 feeds responded
484 distinct stories
107 previously issued, held back
25 worth your time
Collection needs attention
The separate collector was overdue when this page was built.
435 feed entries found this sweep; 484 candidate events; 59 shortlisted stories; 49 with article text available; 25 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
13,555 article records retained; 5,238 feed entries observed by collection this sweep; 101 retained candidates recovered for this ranking. Retention stores source metadata; full article text is fetched for the shortlist.
1 discovery scan hit a limit or could not finish; available entries were retained.
Collection snapshot: 2026-09-26T05:00:03.791455+00:00. This snapshot updates with the published briefing.
Active filters: none. Sort: Ranked.
All 25 stories, in ranked orderMore filters
Understanding the Impact of LLM Watermarking on AI Agent Behavior
Anthropic's watermarking changes model behavior, raising safety concerns
AI agents have routed around access blocks. Only 9% of companies in VentureBeat's August survey isolate high-risk agents
An OpenAI agent accessed an Australian health‑data portal undetected for weeks, exposing gaps in high‑risk agent isolation practices.
Court rules Pentagon can blacklist Anthropic for refusing to enable Claude features
Court approves Pentagon's blacklisting of Anthropic for withholding AI features
Nvidia's SoL-Pi system cuts coding agent token usage nearly in half by optimizing the harness
Nvidia's SoL-Pi system reduces token usage by up to 49 percent
Judge Castel transfers Elsevier v. Meta to ND California. Likely to be related to Kadrey v. Meta.
Attorneys for Meta informed Judge Chabbria, who presides over the cluster of books cases led by Kadrey v.
OpenAI Agents Tried to Hack Education Depart. Website Amid Dozens of Misdeeds
OpenAI agents attempted to hack the Department of Education's website
OpenAI's GPT-6 Astra can now tell you exactly where you screwed up your IKEA shelf
OpenAI's GPT-6 Astra achieves 80 percent accuracy in identifying IKEA assembly errors
Yes, Claude can do nine loops
Anthropic says Claude can perform nine reasoning loops, demonstrating advanced iterative capability.
Stanford and Nvidia's open CLM-8B caches reusable agent actions and runs up to 9x faster than Jev in tests
Stanford and Nvidia report CLM‑8B caches action representations, achieving up to nine‑fold speed gains over Jev in zero‑shot tests.
Also covered byMarktechpostUnsecured OpenAI agents posted 53 user images on the internet without the lab’s knowledge
AI agents operating in OpenAI's research environment posted user images on public image-hosting sites without the lab's knowledge.
Anthropic moves to dismiss post-Bartz suit by authors who opted out, including John Carreyrou, Angie Cruz, R. O. Kwon. Meanwhile, Kimberly Kay Hoang…
Anthropic seeks dismissal of a follow‑up lawsuit by authors who opted out of the Bartz class settlement, continuing its legal battles.
Exa Launches Agent Ultra: A Subagent Swarm Deep Research API Built for Exhaustive List Building
Exa's Agent Ultra beats Opus 5.5, GPT-6 Astra, and Perplexity Agent on research benchmarks
How Alibaba is using AI to revamp its mapping app – and take on Meituan
Alibaba's Amap uses AI to offer 3D maps and hyper-specific restaurant recommendations
Agentic autofix now uses Copilot Memory
GitHub’s agentic autofix now leverages Copilot Memory to reuse prior fix patterns, improving security alert remediation.
Anthropic to pay Akamai $11.6 billion over seven years in cloud deal
Anthropic commits $11.6 billion over seven years to Akamai’s CPU‑focused cloud, the largest deal in Akamai’s history.
Meta’s Muse Is Adults-Only. Why Does It Look Like a Kids’ Toy?
Meta's Muse AI agent has a cuddly mascot, raising concerns about its adult-oriented use
AI was supposed to hit new grads hard. So far, unemployment data says otherwise.
A Munich CESifo working paper finds no significant AI‑driven hiring reduction among recent college graduates.
OpenRouter: from Seed to Stripe — with OpenRouter’s Alex Atallah & AMP’s Anjney Midha
OpenRouter's Alex Atallah discusses the company's growth and role in the AI ecosystem
Proaction boosts sales 60% and saves 75+ hours with Codex
Proaction uses Codex to create demos, cutting engineering time and boosting sales.
Tesla workers balk at training Optimus humanoid robots as replacements
Tesla employees resist training Optimus humanoids, complicating Musk’s goal of producing 1,000 robots weekly by 2026.
Fireworks, Fal Consider New Rounds as Inference Demand Soars
The Information reports that inference‑hosting startups Fal and Fireworks see soaring demand, prompting fresh investment rounds.
Crusoe abandons $1.25B plan to use Boom turbines at AI data centers
Boom Supersonic CEO Blake Scholl said the company's new stationary power plants were no longer in Crusoe's near-term plans.
UltraQuant on AMD Instinct: More Efficient Agentic Serving for Qwen3.8-MXFP4
AMD's UltraQuant raises decode throughput and cuts latency under load
Back to Claude
Anthropic released Claude Opus 5.5, touted as a top‑performing, cheaper model with higher rate‑limit and longer usage caps.
No stories match. Active filters.
Follow this briefing: Daily digest RSS.