AI Slashes Bug Hunting From Nine Months to 90 Hours, and Attackers Can Use It Too
Security firm Forescout has published a report, Finding Critical Infrastructure Bugs with Off-the-Shelf AI, showing how publicly available AI models can speed up the search for software flaws. Researchers built a custom AI-assisted harness and used it to examine 10 popular TCP/IP stacks, the software that lets devices communicate over networks.
The results were striking. Analysis time fell from around nine months to 90 hours, and the team uncovered 99 vulnerabilities, including 66 zero days (flaws not previously known). The AI analysis itself took about 10 hours, after which researchers spent roughly 80 hours manually checking the findings. Static analysis across the 10 stacks cost around $935, or $2.58 per finding, according to the report. Analysing compiled binaries rather than source code was around six times more expensive and took twice as long.
Forescout was clear that AI did not replace human expertise. Daniel dos Santos, its VP of Research, said the technology dramatically compressed the work, and that the same speed and low cost are available to attackers as well as researchers. He said this is especially significant for critical infrastructure, where older and hard-to-patch devices can stay in service for years.
The picture is not one-sided. ReliaQuest research suggests AI helps threat actors make attacks cheaper, faster to scale, easier to customise and harder to spot. Others have cautioned that AI can make attacks less efficient, and a Proofpoint expert recently described AI-generated exploits as highly flawed.
Why it matters
The cost of finding bugs in embedded and network software has dropped sharply, so the assumption that obscure or older devices are too costly for attackers to study no longer holds. Organisations running legacy or hard-to-patch equipment are most exposed, because the time between a flaw being found and being exploited may shrink while their ability to patch stays slow. This forces a decision on compensating controls, such as isolating devices that cannot be updated.
What to do
- Inventory connected and embedded devices that rely on third-party TCP/IP stacks.
- Identify legacy devices that cannot be patched and isolate them from critical networks.
- Shorten the time taken to prioritise and apply vendor security patches.