Anthropic Says AI Worms May Be Months Away—But the Warning Faces a PR Reality Check
Anthropic Says AI Worms May Be Months Away—But the Warning Faces a PR Reality Check
Anthropic’s warning begins with a familiar cybersecurity threat and pushes it into new territory: a worm that does not merely copy code, but carries an AI model with it. The company says small models are now approaching the capabilities needed to move across insecure internet-connected devices, from cheap Android TV boxes to smart appliances.
The immediate evidence is narrower than the headline scenario. Researchers linked 15 off-brand Android TV boxes, costing $55 each, and ran Google’s 26-billion-parameter Gemma model across the cluster. The experiment showed that consumer hardware can collectively supply an alternative to data-centre compute; it did not produce a fully autonomous rogue replicator.
Still, Anthropic argues the trajectory matters. Models with 30 billion parameters or fewer can already perform many parts of the replication chain, and the researchers predict that, at historical rates of improvement, small models could reach the relevant capability level by the end of 2026. “We don’t yet have a full rogue replicator,” they wrote, but agents are increasingly able to complete the separate steps and pass adaptations to later copies.
That is what makes the prospect unsettling: there may be no single server to switch off. Anthropic says a future worm could be a slow, distributed “digital slime mold” living on insecure devices—or a faster cyber system targeting more valuable infrastructure. The lab’s position is not that this outcome is inevitable, but that security teams should harden devices before autonomous agents can exploit them at scale.
The warning has also drawn a political and rhetorical backlash. David Sacks amplified a post dismissing the announcement’s framing as more dramatic than reality, arguing that the underlying research deserves a cooler reading. The disagreement is therefore less about whether IoT security is weak than about timing: Anthropic sees early warning signs; its critics see a speculative leap from a controlled demonstration to household AI contagion.
Write a comment