Anthropic's threat report shows the safeguards are doing exactly what they should. Every single operation — Russian espionage crews, ShinyHunters extortionists, hacktivists running stolen API keys — was detected, disrupted and reported to authorities and industry partners. Publishing the playbook openly lets defenders everywhere spot the same activity, and that transparency is worth more than panic.
Catching a few plots after the fact is not a safety record. Scientists tied to state programs slipped past regional blocks and disguised research into chikungunya, avian flu and toxin design, and the company admitted it could not reliably separate legitimate work from weapons development. Nothing short of hard restrictions on the highest-risk capabilities and real international rules will do here.
© 2026 Improve the News Foundation.
All rights reserved.
Version 7.4.1