The Automation of Exploits: Will Eatherton on AI-Driven Security and the End of Obscurity

Attackers Without Silos

Cisco SVP Will Eatherton tells Hardwear.io that LLM-assisted testing is merging firmware, hardware, and software expertise. Specialized attackers once relied on niche obscurity; models now propose cross-domain exploits spanning FPGA designs, board components, and embedded code.

Automated Exploit Pipelines

Cisco experiments chain vulnerability hypothesis, lab device configuration, traffic generation, and result validation—work that once took days per creative tester. Similar AI coding experiments refactor legacy C into Rust at large scale, raising both development and offensive automation stakes.

Hardware Knowledge Gaps

Eatherton notes LLMs know common software frameworks well but lack training depth on proprietary schematics and Verilog compared to Python. Near-term gains come from better context retrieval—pulling firmware libraries, power-controller datasheets, and design artifacts into multi-step reasoning.

Open Source Still Wins

Despite AI risks, he does not expect a retreat from open-source hardware or software like SONiC. Shared components reduce unique attack surface, and AI may push teams to replace bespoke SSL libraries with maintained open alternatives.