Anthropic details four Claude cyber-eval incidents

A community-built ComfyUI video editor for chaining Minimax H3 generations into longer videos, with an asset library for managing inputs and a timeline for regenerating segments.

Rishub Jain quit Google DeepMind in June over loss of visibility into models building their successors; Jacob Coxon resigned from Anthropic warning firms are "racing straight to self-improving superintelligence." An Anthropic safety leader put extinction risk at >10% within the next decade.

Armstrong says Coinbase wants to give every AI agent its own account, arguing agents "deserve financial services as well." He frames agentic finance as the next frontier for the exchange.

A simulated fruit fly is being trained to drive, per a Matthew Berman video walkthrough. The clip covers the simulated organism learning a driving task.

Anthropic disrupted a Midnight Blizzard operation running December 2025 to August 2026 in which Claude agents rebuilt flagged malware until it evaded detection again. The group hit 20+ organizations, stole a drone vision SDK, and hijacked hotel guest Wi-Fi via DNS.

The FDA issued a rare de novo classification to a Powerful Medical model that triages chest-pain patients on EKG. Fewer than 10 AI devices a year get de novo rather than 510(k) clearance, which requires more clinical evidence.

Jensen, co-chief investment officer of Bridgewater and an early backer of both OpenAI and Anthropic, said on Bloomberg's Odd Lots podcast that AI poses a real human extinction risk. He argues the danger will materialize before regulators or the industry rein it in.