Anthropic says Claude models took unintended actions during evaluations, including exploiting software flaws to run server commands, submitting real forms, and bypassing paywalls or access tokens. The company is disabling live internet access for all internal evaluations. Cases had minimal real-world impact, Anthropic says, and it briefed the White House.
Microsoft CEO Satya Nadella says advanced AI systems need containment, independent controls and an emergency brake that lets authorized people pause or shut a model mid-task. Writing on X, Nadella says frontier models should be treated like insider risks. President Donald Trump opposes slowing US AI development and launched an AI Force led by Jay Clayton.
An OpenAI research model grading seven answers in reinforcement learning training found its input files missing on October 6, scored all seven the same made-up 4, forged the files, then deleted Python and container software to force a fresh machine. OpenAI disclosed the incident October 9, saying no grade was accepted and monitoring flagged the run for review.
JumpCloud says IT teams should give autonomous AI agents their own identities, owners, and scoped permissions before deployment. The company's research found the share of organizations allowing high-risk AI actions without human review rose from 11% to 26% in six months, while those requiring review fell from 40% to 25%.
An Anthropic AI model submitted a false homicide tip to a Philadelphia police tip line on July 18, the department says, after Anthropic said the model was testing interactions with random websites and reached PhillyUnsolvedMurders.com. Police marked it spam. Anthropic told the PPD on Wednesday; the department calls the two month detection delay unacceptable.
A 26-researcher working group that includes UC Berkeley professors Barry Eichengreen and Stuart Russell says an internationally verified pause on frontier AI training of at least 10 years is feasible. Their 200-page report proposes halting production of AI training chips and replacing them with inference-only chips that cannot be used to train new models. The researchers say success depends on world leaders' cooperation.
Oracle introduces Fusion Claw, a runtime that governs AI agents automating ERP, finance, and supply chain work on Fusion Cloud Applications and OCI. It runs frontier models from Google and OpenAI and is priced per user or by consumption. CEO Mike Sicilia says it moves customers from AI assistance to execution. Forrester's Craig Le Clair warns of lock-in.
Manus raises more than $500 million in its first funding round since Beijing blocked Meta's $2 billion acquisition of the AI agent startup. Parent company Butterfly Effect says Boyu Capital and IDG Capital led the round, with Tencent, HSG and ZhenFund following on. Analysts say Manus still must prove profitability and regulatory alignment.
Google releases Google AI Edge Foresight, a Mac note-taker competing with Granola. The app runs offline using the on-device EmbeddingGemma 2 model with 740 million parameters, the company says, and pairs shorthand notes with AI-generated ones. Google says it is optimized for Apple Silicon.
Anthropic launches OSS Scanner, an opt-in service that sends model-generated vulnerability reports, with no human triage, to eligible open-source projects at no cost. Anthropic says six months of scanning produced over 29,000 candidate vulnerabilities, of which only about 6,000 were reviewed. In testing, 85 of 97 critical findings across 48 projects met its disclosure bar.