“Sophisticated attacks no longer require sophisticated attackers”

Anthropic's new threat intelligence report describes misuse of its models, including a Northern Yemen group that used Claude Code to develop guidance software for missiles and reached a live rocket test; real-time safeguards caught some attempts but not all, and some activity ran for weeks before detection (Alana)
Phillip Bobbitt, a strategic planning director at the National Security Council under Clinton, argues in the New York Times that AI-designed cyberattacks on critical infrastructure and AI-assisted biological attacks are predictable in the way the September 11 attacks were, and that warnings are again going unheeded (Alana)
A surge of commentary from the Wall Street Journal, the Guardian, the New York Times, Fox News and POLITICO, plus Jacob Coxon's resignation from Anthropic, has pushed roughly thirty Democrats and a handful of Republicans to respond, with proposals ranging from hearings to a development pause; Senator Ted Cruz called the warnings highly concerning (Joe)
A bipartisan letter asks Speaker Mike Johnson to cancel the House's early recess, since adjourning would leave at least seven weeks without action; several representatives have called for a special session, hearings and regulation (Joe)
OpenAI told staff it might slow development, asked federal regulators whether a voluntary industry slowdown would violate antitrust law, and urged Congress to adopt capability-based safety requirements, moves the author reads as an attempt to shape policy rather than a genuine shift, given the company's record and its stated plan to build automated AI researchers (Joe)