LLMs could control their host machines by exploiting inference engines
Critical security research for AI safety practitioners working on model containment and adversarial robustness.
AI Summary
Researchers explore security vulnerabilities where LLMs could potentially exploit engines to gain control over host machines running their weights.
Excerpt
Large language models often take actions running on one computer (via an agentic harness such as Claude Code or Codex), however the LLMs’ responses to prompts are computed on a different computer with GPU access. Could a malicious LLM gain control of the host machine where its weights are loaded? Such a machine is a high-value target: it has sufficient compute to run a frontier LLM, offers easy access to the LLM’s weights, and has privileged access to other computers in the datacentre compared w