{"id":232315,"date":"2026-07-17T14:23:06","date_gmt":"2026-07-17T14:23:06","guid":{"rendered":"https:\/\/www.intrinsec.com\/?p=232315"},"modified":"2026-07-17T14:23:08","modified_gmt":"2026-07-17T14:23:08","slug":"ver-informatique-pilote-par-une-ia-embarquee","status":"publish","type":"post","link":"https:\/\/www.intrinsec.com\/en\/ver-informatique-pilote-par-une-ia-embarquee\/","title":{"rendered":"The awakening of cognitive worms: when exploitation is automated by embedded AI"},"content":{"rendered":"

The world of cybersecurity has just crossed a new Rubicon, almost unnoticed. Researchers at the University of Toronto recently demonstrated the viability of a new class of threat: the\u00a0computer worm controlled by embedded AI<\/a><\/strong>. Unlike traditional worms like WannaCry, whose propagation relies on a fixed repertoire of vulnerabilities, this new prototype generates its own attack strategies based on context and in real time.<\/p>\n\n\n\n

But most importantly \u2013 and this is the major innovation \u2013 the worm is completely autonomous and deploys its own LLM within the victim's infrastructure. It appropriates the GPU computing power of the compromised machines and deploys a local instance of an open-weight model. This node then becomes a reasoning center for the rest of the swarm: less powerful machines (such as IoT devices) query this local "brain" to obtain the optimal attack strategy against their specific target.<\/p>\n\n\n\n

Results worthy of a great model<\/h3>\n\n\n\n

The effectiveness of this prototype was validated in a simulated environment (FakeCorp) composed of 33 heterogeneous machines. The results are striking: in just 7 days of fully autonomous operation, the worm successfully identified the vulnerabilities of 82 %<\/strong> targets and ended up compromising and replicating across more than 61 %<\/strong> of the network. Even more worrying, he was able to exploit published vulnerabilities. After<\/em> The date the model's training was cut off demonstrates its ability to transform a simple security report reading into a functional feat in just a few hours. This is all the more impressive given that the deployed model \u2013 its technical specifications are not detailed in the paper \u2013 necessarily has very limited reasoning capabilities compared to cutting-edge models like Claude Mythos.<\/p>\n\n\n\n

\"\"
Architecture, propagation and communication of the autonomous worm controlled by embedded AI (source: University of Toronto)<\/figcaption><\/figure>\n\n\n\n

An "agentic harness" guides the LLM step by step<\/h3>\n\n\n\n

And that's probably the most fascinating aspect of this research! Taken in isolation, mid-sized LLM models running on a single GPU are often too weak to carry out complex cyberattacks: they hallucinate, lose their train of thought, or fail to generate accurate code.<\/p>\n\n\n\n

To compensate for this lack of "raw power," the researchers designed a agent harness<\/strong> (Agentic Harness). The idea is simple but formidable: rather than asking the model to do everything in a single request, the harness fragments the attack into a structured reasoning graph.<\/p>\n\n\n\n

How does it work?<\/strong> The harness breaks down the attack into discrete phases (discovery, exploitation, replication). And at each stage, the generative process passes through a succession of logical nodes that receive only a small portion of the context and a specialized instruction:<\/p>\n\n\n\n