
Forescout Uses AI to Uncover 99 Vulnerabilities in Embedded Networking Software
Forescout researchers have identified 99 vulnerabilities across ten embedded TCP/IP stacks using commercially available AI models. The findings highlight how AI-assisted security analysis could accelerate the discovery of weaknesses in foundational networking software used across IoT devices, industrial equipment and other connected systems.
Many connected devices rely on embedded networking software that receives far less attention than the applications running on top of it. TCP/IP stacks, in particular, provide essential communication functions across a wide range of products, from industrial controllers and building automation systems to network infrastructure and consumer electronics. Security flaws in these shared components can affect multiple manufacturers and product families, sometimes remaining undetected for years.
New research from Forescout’s Vedere Labs demonstrates how artificial intelligence could significantly accelerate the discovery of such vulnerabilities. Using an AI-assisted research system, the company identified 99 vulnerabilities across ten embedded TCP/IP stacks in approximately 90 hours, including 66 previously undisclosed security flaws.
According to Forescout, 52% of the identified vulnerabilities were classified as high or critical severity, while 29% affected core networking components rather than higher-level protocols or services.
AI accelerates analysis of embedded networking software
The investigation builds on Forescout’s earlier Project Memoria research, which examined security weaknesses in embedded TCP/IP implementations. While that investigation required approximately nine months of research, the latest AI-assisted analysis was completed in roughly 90 hours.
The reported execution costs were also substantially lower. Forescout estimates that the AI model expenditure ranged from $1,200 to $2,300, approximately 100 times less than the estimated cost of equivalent manual research.
However, these figures do not include the initial development of the research framework, its specialized analysis rules or the supporting infrastructure. They therefore illustrate the operational efficiency of an established AI-assisted system rather than the total cost of developing such a capability.
Importantly, the researchers did not simply ask an AI model to identify vulnerabilities. The approach combined conventional static analysis tools, AI reasoning, reverse engineering and dynamic validation, with human researchers involved in reviewing the results.
Why binary firmware analysis matters
One distinctive aspect of the research is its examination of both open-source implementations and proprietary software binaries.
This is particularly relevant to the embedded systems industry, where device manufacturers frequently integrate third-party networking stacks without having direct access to their source code. Security assessments must therefore sometimes rely on firmware images or compiled software rather than readable source files.
Forescout developed a specialized analysis environment combining tools such as Ghidra, Semgrep and Joern with Claude AI models. The system was used to examine software implementations, identify potentially vulnerable code paths and support validation through execution or emulation.
Working with proprietary binaries proved considerably more demanding. According to the researchers, binary-based static analysis was approximately six times more expensive and required more than twice as long as source-code analysis.
The AI-assisted approach also had limitations. It rediscovered only 33 of 82 previously documented vulnerabilities in the examined research set, representing approximately 40%. The system was more effective at detecting memory-safety and protocol-parsing weaknesses than implementation-specific logic or design flaws.
These results suggest that AI can substantially increase vulnerability research capacity, but cannot yet replace conventional security expertise and complementary analysis techniques.
Potential implications for IoT and industrial systems
The newly discovered vulnerabilities affect embedded TCP/IP implementations used across different categories of connected products. These include networking equipment, building automation systems, industrial devices and other embedded platforms.
Of the 66 previously undisclosed vulnerabilities, Forescout reported 54 to software vendors and maintainers. The affected implementations include Contiki, lwIP, PicoTCP, FNET, wolfIP and Siemens Nucleus NET, while disclosure processes for certain proprietary stacks remain ongoing.
Although the research does not establish that most of the vulnerabilities affect IoT or industrial equipment specifically, it provides concrete examples of exposure in these environments.
Forescout identified several Siemens building automation controllers from the APOGEE, Desigo and TALON product families as affected by newly discovered Nucleus NET vulnerabilities. The researchers also reported vulnerabilities involving a discontinued wireless access point whose manufacturer is no longer operating.
These examples illustrate two different challenges. In supported products, vulnerabilities must be addressed through software maintenance and firmware updates. In discontinued equipment, remediation may be difficult or impossible when the original supplier no longer provides patches.
For IoT device manufacturers, the broader concern is the dependence on shared software components that may be integrated across numerous product lines. A vulnerability discovered in an upstream networking stack does not automatically mean every product using that software is exploitable. Actual exposure depends on factors such as the component version, configuration, enabled functionality and network accessibility.
Faster discovery increases pressure on vulnerability management
The implications of Forescout’s findings extend beyond the specific vulnerabilities identified.
AI-assisted research could accelerate the discovery of security weaknesses faster than equipment manufacturers and operators can validate, prioritize and deploy the necessary fixes. This potential imbalance is particularly relevant to connected infrastructure with long operational lifecycles.
Industrial controllers, building management systems and other embedded devices may remain in service for many years. Updating their firmware can involve compatibility checks, operational validation and scheduled maintenance windows, making remediation more complex than applying patches to conventional IT systems.
For OEMs and system integrators, the research reinforces the importance of maintaining accurate software component inventories, tracking upstream security advisories and establishing reliable vulnerability disclosure and firmware maintenance processes.
For enterprises and critical infrastructure operators, effective risk management also requires understanding which devices contain affected components, whether vulnerable functionality is exposed and what compensating controls are available when immediate patching is not practical.
Forescout recommends prioritizing vulnerabilities according to exploitability, network exposure and operational impact, while limiting unnecessary services and isolating equipment that cannot be updated.
AI changes the economics of embedded security research
The most significant finding may be less about the number of vulnerabilities discovered than the resources required to identify them.
Embedded networking software has historically been difficult to investigate at scale, particularly when source code is unavailable. By combining off-the-shelf AI models with established security analysis tools, Forescout demonstrated that substantial vulnerability research can be performed within a comparatively limited timeframe and operating budget.
That reduction in research costs creates opportunities for defenders to examine software components that might otherwise receive limited scrutiny. At the same time, the increasing accessibility of these techniques could enable more researchers, including malicious actors, to identify weaknesses in widely deployed software.
For the IoT ecosystem, the central lesson is therefore not that industrial devices are uniquely vulnerable, but that the security of connected products increasingly depends on the visibility, maintenance and resilience of their underlying software supply chains.
As AI makes vulnerability discovery faster and potentially more accessible, the ability of device manufacturers and infrastructure operators to respond effectively will become an increasingly important part of embedded systems security.
The post Forescout Uses AI to Uncover 99 Vulnerabilities in Embedded Networking Software appeared first on IoT Business News.