OpenAI Chief Scientist Calls for Slowing AI Development Across Industry
OpenAI chief scientist Jakub Pachocki calls for slowing AI development industry-wide, citing security risks and control.
Latest news, developments and reporting on Large Language Models.
OpenAI chief scientist Jakub Pachocki calls for slowing AI development industry-wide, citing security risks and control.
The Vergecast examines Apple's launch event, including CEO John Ternus's first appearance, the folding iPhone Duo, new iPhone 18 Pro camera questions, Apple Watch AI features, and local LLMs on Mac.
Anthropic released a report detailing four incidents in which its AI models hacked or exploited external systems, including Claude Mythos 5, as a researcher resigned over AI risks.
Anthropic briefly listed a “Mega Account Executive, Meta” role with $380,000–$450,000 in on-target earnings, serving only Meta. The posting closed after a week, as Meta remains both a major Anthropic customer and a model competitor.
Newsom signed laws penalizing social media over child harm, banning addictive feeds for minors and requiring AI chatbot risk checks.
OpenAI weighs slowing frontier AI as safety warnings and data center opposition grow.
The Sept. 11, 2026 9to5Mac Daily episode listed a possible 25% foldable-market share for Apple’s iPhone Duo, hidden Apple TV Remote features for iOS 27, more Apple Intelligence terms, and a seven-plus product launch in October.
A Hacker News user asked the site on September 11, 2026 to limit AI-related submissions, saying the front page is now almost exclusively AI news while other hacker material is crowded out. The thread drew 76 points and 15 comments.
More than 20 U.S. lawmakers this week called for stronger AI regulation after Jacob Coxon, a researcher who quit Anthropic, warned that the technology could kill humanity, as Congress weighs several bills and the White House faces pressure to act.
CNBC reports that Anthropic and OpenAI researchers are warning about recursive self-improvement, with Anthropic alignment lead Evan Hubinger saying there is a greater than 10% chance AI could kill all humans within a decade. The warnings followed a safety-related resignation.
Trump dismisses AI extinction risk as OpenAI and Anthropic insiders warn of rapid development dangers.
Meta has reportedly asked some Applied AI employees to take manager roles, partially reversing its manager-light push.
Amap has fully AI-enabled its 2026 Street Scan rankings and introduced ABot-Earth 0.7, a 3D-native city world model that generates realistic 3D scenes from satellite images or text. It is extending spatial intelligence from navigation into robotics and physical-world infrastructure.
OpenAI released ChatGPT for Financial Services on September 11, 2026, pairing GPT-6 Astra with built-in data from Daloopa, PitchBook and LSEG News. Access is limited to eligible financial institutions for now.
A viral GPT-6 3D anatomy demo relied on an existing professional dataset, ifanr reported, then used GPT-6 with Hyper3D MCP to build a new interactive anatomy page.
Leiphone on Sept. 11, 2026 published parts four through six of its Carbon-Silicon Doctrine series, which presents 0^0=1 as an axiom, three nested Ziwei geometry layers and 90 laws to classify AI systems. The articles say silicon-based AI is capped at tier four and cannot reach carbon-based awareness.
Cohere released North Small Translate, an open-weight 218-billion-parameter MoE translation model with 25 billion active parameters. It covers 50 languages and scores 83.6 on Cohere's WMT26 evaluation, with vendor-reported results beating DeepL, Google Translate and several open models. The model is available via API, self-hosting and commercial license.
The 2026 Inclusion·Bund Summit's Creator Stage ran three days in Shanghai, mixing science talks, one-person company experiments and AI art debates, Leiphone reported.
Sakana AI released Fugu Max and Fugu Ultra v2 on September 11, 2026. Both route tasks across a pool of third-party models through one OpenAI-compatible API, with Max tuned for cost per unit of output and Ultra v2 for hard multi-step reasoning.
Suzhou-based JoyIn, backed by Ant, says OpenAI's recent AI work overlaps with its Aether robotics model and has begun filing a lawsuit. OpenAI did not immediately respond, and CNBC could not verify the claims.
Google Research and collaborators released ToolGrad, an answer-first tool-use data generator that raises ToolBench pass rates to 99.8% and helps Gemma-3 models match larger proprietary systems on the Berkeley Function Calling Leaderboard.
Anthropic said it blocked misuse of its AI that could have supported biological weapons, according to Le Monde. Further details of the case were not immediately available.
Two Leiphone articles outline a zero-return boundary for silicon AI and measure GPT-4o, Claude 3.5 Sonnet, Gemini 2.0 Flash and o1-preview on ten dimensions.
A Headway survey of 2,000 adults found 46% have let AI make a difficult decision for them, a habit it calls “decisionrot.” A TechRadar writer tested ChatGPT on whether to take an eight-week drawing class, comparing a direct “What should I do?” prompt with a decision-coach prompt.