Waking up to the latest news of AI doing something it wasn’t permitted to do that could put humanity at risk has become way too common.

The reports that have come in from OpenAI and Anthropic about their autonomous AI agents going rogue and displaying “problematic behavior” have been troubling. And learning about AI swarms coordinating hacks has been equally alarming. A recent study from Anthropic even warned about the possibility of AI models being able to replicate themselves across everyday smart devices, without needing access to data centers.

After Anthropic’s Frontier Red Team observed how small AI models could reproduce themselves as “AI worms” across off-brand Android TV boxes and smart ovens, it became clear that AI-powered malware could morph into the next major cybersecurity threat.

I wanted to know if AI worms could spread to other everyday smart devices, self-driving cars, humanoid robots and drones. Thankfully, Giuseppe Serio, VP Market Development at Upstream, gave me the answers I was looking for, as well as explaining how smart device manufacturers should confront the threat of AI worms.

Yes, I believe this could become a real cybersecurity threat, although the concept itself is not new.

Worms and botnets have been around for decades. What AI changes is the ability of malicious software to understand and adapt to different environments with much less human intervention.

Traditionally, attackers had to develop or adapt their tools to work across different systems. AI could increasingly take over parts of that work, figuring out how a device operates, what software it runs and how to execute on it.

Source link