Skip to content

Latest News

Redwood Researchers Propose Counting Nearly All AI Company Work as Capabilities

Redwood Researchers Propose Counting Nearly All AI Company Work as Capabilities

Nov 18, 2026
redwoodresearch.org

Alek Westover and Nathan Sheffield of Redwood Research propose defining capabilities activities as all AI company work except idle compute, narrow product development, customer serving, and six listed safety tasks. They say false positives beat false negatives when pacing compute. They urge developers to let third-party auditors measure and publicly report the split.

Anthropic Adds Experimental Code Mode to Claude Code for Batched Tool Calls

Anthropic Adds Experimental Code Mode to Claude Code for Batched Tool Calls

Nov 18, 2026
Claude Code Docs

Anthropic adds Code Mode to Claude Code, an experimental feature where Claude writes JavaScript that makes several tool calls in one step instead of asking for each one. It is off by default and requires version 2.1.323 or later, a subscription or Console account, and the CLAUDE_CODE_EXPERIMENTAL_CODE_MODE variable. Code stops after ten minutes unless it runs in the background.

Google Cloud AI Research Releases PaperOrchestra for LaTeX Paper Drafts

Google Cloud AI Research Releases PaperOrchestra for LaTeX Paper Drafts

Nov 18, 2026
GitHub

Google Cloud AI Research publishes PaperOrchestra, an Apache-2.0 framework that turns raw ideas and experimental logs into submission-ready LaTeX manuscripts using agents for outlining, literature search, section writing, refinement and plotting. Authors Yiwen Song, Yale Song, Tomas Pfister and Jinsung Yoon say it targets conference templates. The generation dataset is withheld and will be released later.

Google Threat Intelligence Group Says Google AI Threat Defense Combines Gemini, Wiz, CodeMender, Mandiant

Google Threat Intelligence Group Says Google AI Threat Defense Combines Gemini, Wiz, CodeMender, Mandiant

Nov 18, 2026
Google

Google Threat Intelligence Group says Google AI Threat Defense combines Gemini, Wiz, CodeMender and Mandiant to find and patch flaws continuously. The group's AI Threat Tracker reports attackers use autonomous LLM frameworks to scan software updates and build exploits without human involvement. Chief analyst John Hultquist says attackers still target known, unfixed issues, faster.

GTIG Says Customized PentAGI Ran Recon Against Asian Targets

GTIG Says Customized PentAGI Ran Recon Against Asian Targets

Nov 17, 2026
Google Cloud Blog

Google Threat Intelligence Group and Mandiant report that threat actors now deploy autonomous multi-agent frameworks such as CAIRN and PentAGI in live intrusions, not just manual prompting. GTIG says a customized PentAGI platform ran reconnaissance and scanning against Asian organizations from late August to early October 2026. In LLMJacking cases, Mandiant says single-day fraudulent cloud charges exceeded $100,000.

Page 1 of 622
Next
Showing 1 - 10 of 6216 articles