Forescout Technologies has released findings demonstrating that artificial intelligence can drastically reduce the time required to identify software vulnerabilities. In a report titled Finding Critical Infrastructure Bugs with Off-the-Shelf AI, the company detailed how it utilized publicly available AI models to scan 10 popular TCP/IP stacks. By employing a custom AI-assisted research harness, the team successfully uncovered 99 vulnerabilities, including 66 previously undisclosed zero-day flaws.
The research confirms that AI significantly accelerates the discovery process, compressing a task that traditionally spans nine months into approximately 90 hours. According to the data provided by Forescout, the process involved 10 hours of automated AI analysis followed by 80 hours of manual validation by human experts. The firm noted that while AI does not replace the need for skilled researchers, it changes the economics of vulnerability discovery by allowing complex embedded software to be analyzed at a fraction of the time and cost previously required.
The cost efficiency of the approach was a key finding of the study. Static analysis across the 10 TCP/IP stacks resulted in an expenditure of approximately $935, averaging roughly $2.58 per identified vulnerability. However, the report highlighted a disparity in resource requirements, noting that binary analysis was significantly more resource-intensive, taking twice as long and costing approximately six times more than the analysis of source code.
Forescout emphasized that these advancements in AI-driven research are available to both security teams and malicious actors. This shift in the threat landscape necessitates a more proactive approach to securing critical infrastructure, where legacy devices often remain in operation for extended periods and are difficult to patch. The accelerated pace of vulnerability discovery means that organizations must prioritize risk management more effectively than in the past.
To address the risks posed by this increased speed of discovery, Forescout recommended that security teams move away from treating all vulnerabilities with equal urgency. Instead, organizations should prioritize remediation based on specific factors such as reachability, exploitability, and the potential operational impact of a breach. Furthermore, the firm advised reducing the attack surface by disabling unnecessary services like FTP, Telnet, IPv6, and web administration interfaces on connected devices.
For systems that cannot be patched, such as those that have reached the end of their support lifecycle, the company suggests isolating the devices from the broader network. Additionally, the report advocates for building security into the procurement process by requiring suppliers to provide software bills of materials, publish clear vulnerability disclosure policies, and commit to defined support lifetimes.
While the potential for AI to aid in defensive security is clear, the broader industry landscape remains cautious. Other research has indicated that threat actors are increasingly leveraging AI to scale attacks, lower costs, and customize their campaigns. However, some security analysts have observed that AI-generated exploits can be highly flawed or inefficient, suggesting that the technology is not a universal solution for attackers.
Ultimately, the findings underscore a new reality for cybersecurity professionals: the speed at which vulnerabilities are identified and disclosed is increasing, requiring a corresponding acceleration in patching timelines and defensive strategies. By integrating AI into their workflows, security teams can better keep pace with these evolving threats, provided they maintain rigorous human oversight and focus on risk-based mitigation.





