cyber.hendryadrian.com
Security News · bleepingcomputer.com

Anthropic Warns Small AI Models Could Soon Self-Replicate Across IoT Devices

· Original article ↗

Summary

Anthropic says small AI models may soon be capable of spreading across IoT devices, but its experiments did not demonstrate a fully autonomous worm. The company recommends stronger IoT defenses.

Key points

  • Anthropic tested model deployment and propagation using cheap Android TV boxes; the controlled setup exposed targets with unauthenticated ADB access.
  • Gemma4-26B-A4B ran across 15 Android TV boxes, but the experiments did not demonstrate a fully autonomous self-replicating worm on real IoT hardware.
  • Qwen3.5-9B installed and configured software on eight of nine hardware types tested; GLM-5.2 reached all eight nodes in four of five controlled-network trials.
  • Anthropic estimates models with 30 billion parameters or fewer could reach the required capabilities by the end of 2026, based on results and historical progress.
  • On the studied hardware, model deployment and inference took tens to more than 100 hours, which Anthropic says could give defenders time to respond.
  • Suggested defenses include monitoring large model transfers and unusual device-to-device traffic, using honeypots, quarantining infected subscribers, and improving IoT security defaults.

Article Details

Event Type
Anthropic reported research experiments on whether small AI models could self-replicate across consumer IoT devices; the experiments did not demonstrate a fully autonomous worm.
Impact
A successful self-replicating model could create a distributed botnet that runs AI workloads across consumer devices without centralized infrastructure. Anthropic said the demonstrated process was slow and not an imminent threat, but warned that capability may improve and recommended stronger IoT security measures.

People

Bill ToulasAuthor of the article.Mikko HyppönenNamed as a participant in a security summit promoted in the article.

Vendors

AnthropicAnthropic warns self-replicating AI could spread to your smart devicesGoogle DeepMindAlthough the researchers successfully distributed the Gemma4-26B-A4B model from Google DeepMind on 15 off-brand Android TV boxes, they did not demonstrate a fully autonomous, self-replicating AI worm running on real IoT

Products

Android TVis warning that small AI models could soon be able to self-replicate across consumer IoT devices, from Android TV boxes to smart ovens, creating distributed botnets capable of running AI workloads without relying on

Related Articles