What matters in AI.

Subscribe

Forescout AI found 99 vulnerabilities in device software

Forescout says its AI work was about 90 hours. Manual work in Project Memoria was 9 months.

The security company Forescout says AI models found 99 confirmed vulnerabilities in 10 TCP/IP stacks. A stack is the software that lets devices send data on a network. Sixty-six of the vulnerabilities were new. Of these, 52% were rated high or critical. The software is in industrial equipment, medical devices and network equipment. Siemens found affected building controllers.

How it works

The team built the tool around Claude models and existing security tools. Code scanners flagged patterns that could show a vulnerability. AI agents then checked if an attacker could go to that code. Last, the system ran the software and made a test that triggered the vulnerability.

The numbers

The automated work was 10 hours, and checks by researchers were 80 hours. The model cost was $1,200 to $2,300. The system found again 33 of 82 known vulnerabilities.

What is not known

The tool was weaker on logic vulnerabilities. Software with no source code cost about six times more to check. Other vendors are investigating.

Insight: Low cost does not repair devices that have no vendor

The work cost $1,200 to $2,300. It found vulnerabilities in firmware from a wireless access point company that has closed. Forescout says attackers can use the same method. Its CEO, Barry Mainz, says containing risk, not patching, must come first. This reads as a gap: to find vulnerabilities costs less for all parties, but a patch must come from a vendor. Look for plans to isolate such devices.

Sources

  1. AI uncovers flaws in discontinued devices with no vendor left to patch themhelpnetsecurity.com
AI MATTER · NEWS · AI MATTER · NEWS ·16 NOV2026

Posted

Tags

More in Research

All Research news