AI in the Crosshairs: How Anthropic Stops Bioweapons Wins

Sat Sep 12 2026 02:58:02 GMT+0000 (Coordinated Universal Time)

Anthropic’s new report shows the AI firm has disrupted attempts to weaponise biology, sparking debate with lawmakers and cyber‑criminals.

Anthropic released a threat‑intelligence report that details dozens of attempts to use its AI model, Claude, to create biological weapons and online scams. The firm claims to have blocked or disrupted these projects, a move that fuels warnings from Sen. Bernie Sanders and top AI safety experts. The report lists misuse cases from state‑backed hackers, cyber‑criminal crews, and propaganda groups. It also highlights how similar threats surfaced at Google and other tech giants. The findings add pressure for global AI governance and stronger safeguards.

AI is fighting back against bioweapons schemes

Anthropic, the AI company behind the Claude model, announced it has stopped dozens of projects that tried to use its technology for biological weapons. That’s part of a growing pattern where AI firms publish “threat‑intelligence reports” to show how they block misuse.

The report flags five cases that aimed to support the creation of biological weapons and six that aimed to develop conventional weapon software. In all, the attacks ranged from fake dating apps and hotel Wi‑Fi scams to surveillance of political dissidents.

The warning from Anthropic feeds into a broader debate. U.S. Sen. Bernie Sanders has called for a pause on advanced AI and a ban on “super‑intelligence.” Meanwhile, former OpenAI chief scientist Jakub Pachocki stressed the need for “voluntary slow‑downs” until safeguards are in place.

Anthropic also said it shared intelligence with law‑enforcement agencies and partner firms, helping to spot and block new threats from groups like ShinyHunters, Midnight Blizzard, and state‑backed cyber‑espionage teams.

According to the report, no misuse was seen involving Anthropic’s most powerful model, Mythos‑class, except a single instance of “distillation” – training a cheaper model from the big one.

The company’s findings underscore the fine line between creating useful AI tools and leaving an arsenal in the hands of bad actors.

Key Takeaway: If the same AI can help design vaccines, it can also help design weapons. That’s why early detection and disruption are now part of every A.I. firm’s mission.