Forescout uses Claude-based AI harness to find 99 flaws across 10 TCP/IP stacks, including unpatched discontinued devices
Forescout researchers built a harness around Claude models and existing security tools to test whether generally available AI could find vulnerabilities on a limited budget. They confirmed 99 vulnerabilities across 10 TCP/IP stacks. 66 were previously unknown, and 52% of those were rated high or critical. The affected software is used in industrial, medical and network equipment. Siemens has identified affected building automation controllers in its APOGEE, Desigo and TALON lines, and other vendors are still investigating. This is a research report, and the article names no CVEs and no exploitation in the wild. The main risk is that attackers could use the same low-cost approach, and that some products have no vendor left to patch them.
Entenda a falha
A vulnerabilidade
No CVEs are listed. Forescout confirmed 99 vulnerabilities across 10 TCP/IP stacks, the components that let devices talk over networks. 66 were previously unknown, and 52% of those were rated high or critical. The stacks are used in industrial, medical and network equipment, and most are not named. Siemens has identified affected building automation controllers in its APOGEE, Desigo and TALON lines, and other vendors are still investigating. The articles give no affected or fixed versions, and several projects have released fixes. The flaws work best as memory errors in code that handles network traffic. A discontinued wireless access point also had seven device-specific flaws, including unchangeable administrator credentials and a buffer overflow in the firmware upload process.
Como funciona
The affected code is the part of a TCP/IP stack that receives and parses data arriving over the network. The articles describe the typical memory errors as copying more data than a buffer can hold, or mishandling information received over a network. Data reaches the stack from the network, the stack processes it, and the failure occurs at the step where the data is copied or interpreted. In the access point, the failure happens in the firmware upload process, where a buffer overflow was found. The articles do not detail the exact sequence for each stack.
Por que é vulnerável
The articles say the flaws are mostly memory errors in network-facing code, such as copying more data than a buffer can hold and mishandling data received over a network. There are also logic and design flaws, such as predictable values that should be random, which the tool found harder to detect. The access point has hardcoded administrator credentials that cannot be changed. Many of these bugs went unnoticed because manual review is slow: Project Memoria took about nine months of manual code analysis. Some products also have no vendor left to issue a patch. The articles do not describe what each fix changes.
Como foi descoberta
Forescout researchers built a harness around Claude models and existing security tools to test whether generally available AI could find flaws on a limited budget. Scanners with 32 rules derived from Project Memoria flagged suspect code. AI agents then assessed reachability and exploitability, and a final stage ran the software and built tests that trigger the issue, with researcher oversight. The run produced 362 initial findings and 312 candidates after deduplication, and 99 were confirmed. It rediscovered 33 of 82 known vulnerabilities. Automated analysis took about 10 hours plus about 80 hours of validation, at a model cost of roughly $1,200 to $2,300. Forescout reported 54 new flaws to developers and maintainers, and investigations remain open for three stacks whose names are withheld. The articles report no exploitation in the wild.
PoC existente, privada
O exploit existe, mas quem o tem não o publicou.
The articles do not publish an exploit. They say the pipeline's final stage tries to run the software and produce a test that triggers each issue, so Forescout holds verification-level triggering tests for confirmed flaws. None were released. Because the same low-cost approach is available to attackers, patching and containing exposed and legacy devices is urgent.
Explicação gerada por IA a partir da matéria e dos artigos, em 16/11/2026 04:18.
Detalhes técnicos
- The pipeline starts with code scanners that use 32 rules derived from Project Memoria. AI agents then assess reachability and exploitability. The final stage runs the software and builds triggering tests, with researcher oversight.
- Automated analysis took about 10 hours, followed by about 80 hours of validation. Project Memoria took roughly nine months of manual code analysis.
- Automated code inspection cost $935. Total model costs ranged from about $1,200 to $2,300.
- The run produced 362 initial findings and 312 candidates after deduplication. 99 were confirmed. The system rediscovered 33 of 82 known vulnerabilities.
- It worked best on memory errors in network-facing code. It was weaker on logic and design flaws such as predictable values meant to be random.
- Compiled targets without source code cost about six times as much and took about twice as long.
- Analysis of a discontinued wireless access point from a defunct company found seven device-specific flaws. They included unchangeable administrator credentials and a buffer overflow in the firmware upload process.
Remediação
- Forescout reported 54 of the new vulnerabilities to developers and maintainers, and several projects have issued fixes. Investigations remain open for three stacks whose names are withheld.
- For devices that cannot be patched, Forescout's CEO says the priority should be containing risk rather than patching.
Linha do tempo
-
Forescout uses Claude-based AI harness to find 99 flaws across 10 TCP/IP stacks, including unpatched discontinued devices
ler a análise desta fonte ocultar
Forescout researchers built a harness around Claude models and existing security tools to test whether generally available AI could find vulnerabilities on a limited budget. They confirmed 99 vulnerabilities across 10 TCP/IP stacks. 66 were previously unknown, and 52% of those were rated high or critical. The affected software is used in industrial, medical and network equipment. Siemens has identified affected building automation controllers in its APOGEE, Desigo and TALON lines, and other vendors are still investigating. This is a research report, and the article names no CVEs and no exploitation in the wild. The main risk is that attackers could use the same low-cost approach, and that some products have no vendor left to patch them.
Technical Details
- The pipeline starts with code scanners that use 32 rules derived from Project Memoria. AI agents then assess reachability and exploitability. The final stage runs the software and builds triggering tests, with researcher oversight.
- Automated analysis took about 10 hours, followed by about 80 hours of validation. Project Memoria took roughly nine months of manual code analysis.
- Automated code inspection cost $935. Total model costs ranged from about $1,200 to $2,300.
- The run produced 362 initial findings and 312 candidates after deduplication. 99 were confirmed. The system rediscovered 33 of 82 known vulnerabilities.
- It worked best on memory errors in network-facing code. It was weaker on logic and design flaws such as predictable values meant to be random.
- Compiled targets without source code cost about six times as much and took about twice as long.
- Analysis of a discontinued wireless access point from a defunct company found seven device-specific flaws. They included unchangeable administrator credentials and a buffer overflow in the firmware upload process.
Remediation
- Forescout reported 54 of the new vulnerabilities to developers and maintainers, and several projects have issued fixes. Investigations remain open for three stacks whose names are withheld.
- For devices that cannot be patched, Forescout's CEO says the priority should be containing risk rather than patching.