AI Found Widespread IoT Flaws More Quickly and Cheaply Than Humans Could
CIO Dive, Friday, November 13th, 2026
Forescout's AI harness found 66 new flaws in ten TCP/IP stacks used in IoT and safety-critical devices, quickly and cheaply.
Forescout researchers ran an AI harness built on Claude models against ten popular TCP/IP stacks, including Contiki, Ethernut, lwIP and wolfIP, rediscovering 33 known vulnerabilities and finding 66 new ones, roughly half of them memory-corruption bugs.
The researchers said the harness is good at memory safety bugs in protocol parsing code but misses logic and design flaws.
Twenty-four of the reported bugs have been fixed and maintainers are investigating another 29.
Forescout warned that unsupported legacy products abound and that AI-enhanced research can now cheaply uncover flaws no vendor will patch.