Anthropic: 'AI Worms' Could Start Infecting Devices Like Viruses

Anthropic has warned that malicious 'AI worms' may be just months away from being able to self-replicate and autonomously infect user devices—household appliances and other internet-connected devices.

According to researchers William Zhou and Daniel Freeman from the Frontier Red Team, such systems may begin to deliberately attack everyday devices: Android TVs, smart appliances, cameras, and other internet-connected equipment with basic computing power and weak protection. Small AI models may be 'just months away from being able to self-replicate on commercial IoT devices,' posing a serious threat as they will not have a 'centralized kill switch,' the authors write.

The warning comes amid growing concerns about the increasing mathematical capabilities of AI models, which could eventually break cryptographic signatures used in crypto wallets and threaten about a third of the Bitcoin supply.

Nearly a third of the global Bitcoin volume—6.26 million BTC—is held at addresses vulnerable to AI models or quantum computers. Meanwhile, 4.33 million BTC are at risk due to address reuse, meaning that moving coins to fresh addresses would eliminate this vulnerability, noted Glassnode co-founder Rafael Schultze-Kraft on Thursday on X.

A day earlier, Ethereum researcher Justin Drake warned that rapid AI progress could break the elliptic curve digital signature algorithm (ECDSA) protecting crypto wallets sooner than expected and urged users to transfer funds to fresh addresses. Dragonfly partner Haseeb Qureshi called this warning 'cryptographic pessimism' and stated that moving tokens to new addresses is not a real solution to the problem.