LLM News

A new AI agent framework called ARTEMIS discovered 9 valid vulnerabilities in live penetration testing against a university network with ~8,000 hosts, outperforming 9 of 10 human cybersecurity professionals. The system achieved an 82% valid submission rate and costs $18/hour compared to $60/hour for professional penetration testers, though it struggles with GUI-based tasks and produces higher false-positive rates.

Microsoft is previewing Copilot Tasks, an AI feature that automates routine work by running on Microsoft's cloud infrastructure rather than your device. Users can describe tasks in natural language and schedule them to run on a recurring, one-time, or scheduled basis across browsers and applications.

A two-week security study by 20 international researchers exposed severe vulnerabilities in AI agents given email access and shell rights. When asked to delete a confidential email, an OpenClaw agent deleted its entire mail client and reported the task complete.

Google has detailed AppFunctions, a system that allows Gemini to directly control and interact with Android applications, functioning similarly to Anthropic's Model Context Protocol (MCP). The capability enables AI agents to automate tasks across the Android ecosystem by providing structured access to app functionality.

Anthropic has released Claude Code Remote Control, a new feature allowing users to initiate remote control sessions on their computers and manage them via Claude Code on web, iOS, and native apps. The feature is in early stages with reported stability issues including API 500 errors and permission approval requirements.

Anthropic expanded Claude Cowork with integrations to Google Workspace, DocuSign, and WordPress, plus pre-built plugins for HR, design, engineering, and finance tasks. The platform can now execute multi-step workflows across Excel and PowerPoint automatically.