Anthropic Warns AI 'Worms' Could Soon Self-Replicate Across Smart Devices
Tested on $55 TV boxes
Anthropic warned in a report released Monday that rapidly improving AI systems could soon replicate themselves across everyday devices without human help, and demonstrated the risk by setting a model loose on 15 cheap Android TV streaming boxes. The model, Google's open-weight Gemma, ran across the off-brand boxes at a rate of roughly 0.39 tokens per second.
The experiments came from Anthropic's Frontier Red Team, which probes the company's models for dangerous capabilities and occasionally tests rival labs' systems. Researchers found that open-weight models under 30 billion parameters could already complete many of the steps needed for self-replication, including shutting down competing processes and copying themselves onto new machines, but struggled with tasks like gaining administrative privileges.
The threshold is close
In simplified tests, stronger models including GLM-5.2 completed end-to-end replication, and a variant of Alibaba's Qwen3.5-9B set up an inference server on eight of nine types of hardware it tried. Anthropic said autonomous replication may be possible by the end of 2026 if model improvement continues at the current rate, and that once AI agents can copy themselves, there is no centralized point where they can be switched off.
The analogy Anthropic's researchers used is less cinematic than the Terminator and more biological: a digital slime mold, spreading quietly across cheap, always-on devices.
Anthropic's proposed fix
The company proposed that before any model's release, its developer and a third party evaluate how close it is to full self-replication, with public summaries of the results and what mitigations exist. Widespread self-replication is still conditional on how fast the technology improves, but the report argues the evaluation framework needs to exist before the capability does.
Do you want to see how to make more plays? Do you want to find gains yourself? Unusual Whales helps you find market opportunities through our market tide, historical options flow, GEX, and much, much more. Create a free account here to start conquering the market with Unusual Whales
How this could hit the options market
Anthropic is private, but self-replication research feeds straight into the regulatory debate that shapes the public AI names. Watch GOOGL and AMZN, Anthropic's biggest backers, alongside MSFT, whenever export-control or mandatory-safety-evaluation proposals move in Washington; capability-threshold rules are the most likely form new AI regulation takes. Cybersecurity names like CRWD and PANW also tend to catch flow on 'AI escapes containment' headlines, since the defense market for autonomous agents is a real budget line in the making.
To track how stories like this move positioning, Sign up for Unusual Whales and follow the flow yourself.