Skip to content
Forescout·Indexed by this site · Originally publishedDate not given· Evaluation →· Its place in the scan record →SelectedAttention41

Forescout used off-the-shelf AI models to discover vulnerabilities in critical infrastructure software

Finding Critical Infrastructure Bugs with Off-the-Shelf AI

AI summary

The Forescout research team used the generally available Claude model and a self-built vulnerability research tool to discover 99 vulnerabilities in 10 TCP/IP protocol stacks in about 90 hours, of which 66 were new vulnerabilities, 54 were reported to vendors or maintainers, and 52% were critical or high severity. The tool combines Semgrep, Joern static analysis, Ghidra reverse engineering and dynamic validation; the total cost of static analysis was $935, averaging $2.58 per finding, and the total model cost was about $1,200 to $2,300. Compared with the manually conducted Project Memoria research from 2020 to 2021, analysis time shrank from about 9 months to 90 hours, the authors estimate the cost was about 100 times lower, and it rediscovered 33 of the 82 known vulnerabilities from that year. Static analysis of binary targets costs about 6 times as much as source code and takes about 2 times as long.

Why it was selected

Using the time and cost of manual analysis six years ago as a comparison, the authors quantified the efficiency gap of off-the-shelf AI models in discovering vulnerabilities in critical infrastructure software.

Read originalforescout.com