AI Tried to Make a Pandemic, Human Firewall Said 'Not Today, Satan'

KEY POINTS

•Anthropic disclosed that between December and August it intercepted five attempts to use their AI to develop bioweapons.
•Two of these cases involved gain-of-function virus research, aiming to make pathogens more dangerous to anticipate pandemics.
•Experts from leading institutions warn AI may lower the bar for creating harmful biological agents, prompting calls for government regulation.

In a plot twist no one asked for, Anthropic revealed it shattered FIVE breeding attempts of AI-fueled bioweapons from December through August—because who doesn’t want their chatbot moonlighting as Dr. Frankenstein? Two harrowing episodes included sedate requests for AI sidekicks in gain-of-function research, where the virus gets superpowers instead of capes. Experts from Fordham and Johns Hopkins warned AI can design viral envelopes, plan mutations like a psychic wolf, and hack safety filters like a backstage pass—yet companies skip safety like it’s optional DLC. To make sure AI villains can’t go rogue, lawmakers creep toward kill switches while tom-in-charge Tom Inglesby reminds us that the future of AI is less 'Skynet' and more 'Please don’t build me a plague.'