FireFlash: open recipe for long-horizon agentic RL

Cursor now scaffolds a web app, spins up a Postgres database, sets up auth behind Cursor SSO and provisions hosting, returning a live link on a chosen domain. Publishing runs deterministic and agentic security checks; nothing goes live until they pass and the user explicitly publishes.

Led by Associate Attorney General Stanley Woodward and Associate Deputy Attorney General Stephen Weiss, the DOJ-SITF targets AI misuse by foreign governments, criminal groups and fraudsters. It also pushes AI developers to tighten internal controls so systems "behave as intended."

James Gates of Charlotte alleges his wife used ChatGPT to get instructions for poisoning him, after he found a handwritten note listing substances and dosage plans, some labeled "lethal." The suit follows her complaints to the chatbot about their relationship.

METR's October 2026 assessment estimates Anthropic is improving model capabilities about twice as fast as before, with benchmark data from unreleased internal models pushing its acceleration estimate to 1.9x versus the 1.7x in Haiku 5.5's system card. METR found no clear bottlenecks preventing further acceleration.

DDEX, the not-for-profit standards-setting organization for music-industry data exchange, admitted ElevenLabs as a full member — its first AI company. The AI audio research firm joins the body that maintains metadata standards used across music distribution and publishing.