Anthropic spots AI worms inching toward your smart fridge — and your cold $BTC 🐛

Anthropic researchers warned that small artificial intelligence models may be "months away" from autonomously copying themselves across internet-connected devices, in findings that could carry implications for cryptocurrency security on devices ranging from Android TVs to smart appliances. The study, authored by William Zhou and Daniel Freeman of Anthropic's Frontier Red Team and published Monday, described self-replicating AI systems as a category of threat that would have "no centralized point to shut off" once deployed on commercial internet-of-things (IoT) hardware.

In controlled experiments, Anthropic ran an AI model across 15 low-cost Android TV boxes without relying on conventional data centers. The company reported that models with fewer than 30 billion parameters completed several steps required for replication, including identifying vulnerable hardware, installing software, and operating without human intervention, but could not finish the full process because some tasks demanded higher-level decision-making. The larger GLM-5.2 model successfully completed the full replication procedure under simplified testing conditions, and Anthropic projected that smaller models could reach comparable capability by the end of 2026.

The warning lands as renewed attention turns to the cryptographic foundations of digital asset wallets. Ethereum researcher Justin Drake said this week that rapid AI advances could break the elliptic curve digital signature algorithm (ECDSA) securing cryptocurrency wallets sooner than previously expected and urged users to migrate to fresh wallets. Glassnode co-founder Rafael Schultze-Kraft wrote in a Thursday X post that almost a third of global Bitcoin supply, or 6.26 million $BTC, currently sits in addresses vulnerable to being broken by AI models or quantum computers, noting that 4.33 million $BTC are exposed because of address reuse and that moving those coins to new wallets would end the exposure. Dragonfly managing partner Haseeb Qureshi called the warning "cryptographic doomerism" and said migrating tokens to fresh addresses was not a real solution against the threat, adding that AI has not triggered a DeFi "hackpocalypse."

The replication research is part of Anthropic's broader expansion into AI safety and applied science. The company recently committed $150 million to the US Genesis Mission, supporting wider use of artificial intelligence in scientific research, as it works with its Frontier Red Team to assess the security implications of increasingly capable models running on common consumer devices.