AI finds 99 flaws in critical infrastructure software in 90 hours, Forescout reports
By aminah
·17 Nov 2026
·4 min read

Off-the-shelf artificial intelligence models found 99 vulnerabilities, including 66 previously unknown flaws, in the networking software that underpins critical infrastructure, in about 90 hours of work that would have taken around nine months manually, Forescout Research has reported.
The findings come from Forescout Research - Vedere Labs and were announced on 12 November 2026 in a report titled "Finding Critical Infrastructure Bugs with Off-the-Shelf AI". The researchers built a specialised AI-assisted vulnerability research harness and applied it to 10 popular TCP/IP stacks, the foundational software that lets connected devices communicate over a network, including seven stacks Vedere Labs had previously analysed and three new ones.
Of the 99 validated vulnerabilities, 33 were previously known flaws that the AI system rediscovered, and 66 were new. More than half of the new flaws, 52 per cent, were rated high or critical severity, and 54 of them were reported to vendors and maintainers. Responsible disclosure remains ongoing for some findings.
## Nine months of work in 90 hours
The time comparison is the headline. Forescout says the AI-assisted approach reduced analysis time from approximately nine months to 90 hours, combining roughly 10 hours of independent AI analysis with about 80 hours of human validation. The automated static analysis cost just 935 US dollars, averaging 2.58 dollars per finding, although total model costs were reported at roughly 1,200 to 2,300 dollars.
"AI did not eliminate the need for expert vulnerability researchers, but it dramatically compressed the work," said Daniel dos Santos, VP of Research at Forescout. "With access to an effective AI-assisted workflow, vulnerabilities in complex embedded software can be found much faster and at a fraction of the cost, by security researchers and attackers alike."
That last point is the warning at the heart of the report. The same cheap, fast techniques that help defenders find flaws can be used by attackers, and Forescout says its research demonstrates that AI is "changing the economics of vulnerability discovery and accelerating the pace at which security teams need to identify, prioritize, and mitigate cyber risk".
## Why TCP/IP stacks matter
TCP/IP stacks are embedded in products used across operational technology, the internet of things, medical devices, network equipment and other cyber-physical systems. Because the same underlying software can be used across products from many manufacturers, a single vulnerability can create risk across the broader technology supply chain.
Siemens was the first vendor to confirm affected devices, identifying building automation controllers in its APOGEE, Desigo and TALON product lines. One published example from the research is CVE-2026-120596, a remote denial-of-service flaw in the Ethernut TCP/IP stack.
The risk is particularly difficult to address when vulnerable software is embedded in devices that may remain in service for decades, even after vendor support ends. Even when a fix is available, it must travel from the software developer through firmware and product vendors before it reaches the organisation operating the device. For discontinued or unsupported products, a patch may never come.
## What organisations should do
Forescout recommends that organisations assess and prioritise risk based on reachability, exploitability and operational impact, rather than treating every vulnerability equally. It advises reducing unnecessary exposure of connected devices, disabling or restricting services such as FTP, Telnet, IPv6 and web administration interfaces where possible, and isolating devices that cannot be patched in tightly controlled network segments.
Patching timelines should be accelerated where fixes exist, Forescout says, and procurement should require suppliers to define support lifetimes, publish vulnerability disclosure policies and provide software bills of materials.
"The speed of vulnerability discovery is increasing with AI, but the reality for asset owners and operators hasn't changed," said Barry Mainz, CEO of Forescout. "Organizations still have legacy systems, operational requirements, and devices that can't easily be patched or simply taken offline. Containing risk, rather than patching, must be the priority."
The study also had limits. The AI system rediscovered only 40 per cent of previously known vulnerabilities in the stacks it was tested against, and it was weaker on logic and design flaws than on implementation bugs. The 99 findings are researcher-validated vulnerabilities, not evidence of 99 active attacks, and Forescout reports no exploitation of the new flaws.
## FAQ
**How many vulnerabilities did the AI find?**
Researchers validated 99 vulnerabilities across 10 TCP/IP stacks: 66 previously unknown flaws and 33 previously known flaws rediscovered by the system.
**How much did it cost?**
Automated static analysis cost about 935 US dollars, or 2.58 dollars per finding. Total model spending was reported at roughly 1,200 to 2,300 dollars.
**Does this mean the flaws are being exploited?**
No. Forescout reports validated vulnerabilities, not attacks. The research shows how quickly flaws can now be found, which is why the company urges faster patching and better isolation of devices that cannot be patched.
Sources: Forescout, "Forescout Research Shows AI Cut Vulnerability Hunting from Nine Months to 90 Hours"; Que, "AI Slashes Vulnerability Discovery Time in Critical Infrastructure" (https://que.com/ai-slashes-vulnerability-discovery-time-in-critical-infrastructure/).