📰 Key Takeaways

Anthropic’s Frontier Red Team researchers William Zhou and Daniel Freeman published a report on Monday warning that self-replicating “AI worms” could become capable of attacks within months, able to automatically infect smart TVs, smart appliances, cameras, and other IoT devices that have limited computing power and weak security, with the entire process requiring no human involvement. The report notes that small AI models may be “just months away” from being able to self-replicate on consumer-grade IoT devices, and once that happens, the threat would be particularly difficult to deal with since there’s no single centralized control point to shut it down.

This warning echoes a separate wave of concern about AI’s growing math capabilities potentially cracking the signature algorithms that secure crypto wallets. According to Glassnode co-founder Rafael Schultze-Kraft, who posted on X on Thursday, roughly one-third of the global bitcoin supply (6.26 million BTC) currently sits in addresses that could potentially be cracked by AI or quantum computers, with 4.33 million of those exposed due to address reuse — a risk that can be eliminated simply by moving funds to a new wallet. Separately, Ethereum researcher Justin Drake warned on Wednesday that AI progress could break the ECDSA signature algorithm protecting crypto wallets faster than expected, urging users to migrate to new wallets as soon as possible. However, Dragonfly managing partner Haseeb Qureshi pushed back on this framing, calling it “cryptography doomerism” and arguing that moving tokens to a new address doesn’t actually solve the underlying threat.

💬 JudyAI Lab Take

This week, researchers at Anthropic’s Frontier Red Team warned that small AI models could be just months away from being able to self-replicate on IoT devices like smart TVs and cameras — forming “AI worms” with no single control point that can be shut down. This is a risk worth flagging early for anyone of us building with AI.

This warning lines up with a separate round of concern in the crypto world. Glassnode’s co-founder points out that roughly a third of the global bitcoin supply sits in addresses that could potentially be cracked by AI, and an Ethereum researcher is warning that the ECDSA signature algorithm could get broken faster than expected — though others argue that’s just “cryptography doomerism” and that simply migrating addresses isn’t a real fix. What stands out to us across both stories is the same pattern: the pace of AI capability gains keeps outrunning the defensive side’s readiness, whether that’s firmware update cycles for IoT devices or how willing people are to actually migrate off old wallet addresses — defenders consistently move slower than the attack surface expands.

The real question for us isn’t whether AI will get misused — it’s whether the systems we maintain actually have a kill switch we can reach when it matters.

📅 Source Info

🔗 Further Reading