Petronella.ai

AI Vulnerability Discovery Cuts Analysis to 90 Hours

November 16, 2026 · Cybersecurity
AI Vulnerability Discovery Cuts Analysis to 90 Hours

Forescout researchers used AI vulnerability discovery to cut analysis of 10 TCP/IP stacks from roughly nine months to 90 hours, finding 66 zero days, a speed shift defenders must plan for.

The findings come from a new Forescout Technologies report, Finding Critical Infrastructure Bugs with Off-the-Shelf AI, covered by Infosecurity Magazine. The report details how publicly available models helped a small-budget team unearth vulnerabilities in open source software and proprietary binaries. Anyone responsible for industrial control systems, embedded devices, medical equipment, or any long-lived connected technology should read this closely: the economics of finding bugs just changed for researchers and attackers alike.

Key Takeaways

What Happened: The Facts in Detail

The report and the team behind it

Forescout Technologies published a report titled Finding Critical Infrastructure Bugs with Off-the-Shelf AI. Its stated purpose was to show how publicly available AI models can help teams working with a small budget find vulnerabilities in open source software and proprietary binaries. The company has been adding its voice to a growing number of vendors urging security teams to embrace AI, and this report is its evidence for what that looks like in practice on the offensive research side.

The test: 10 TCP/IP stacks under an AI-assisted harness

The researchers did not test a single product. They built a specially constructed AI-assisted vulnerability research harness and pointed it at 10 popular TCP/IP stacks. That choice is significant for regulated businesses. TCP/IP stacks are the networking code embedded in operating systems, firmware, and connected devices of every kind. A flaw in one of these stacks can affect thousands of downstream products at once, which is exactly the pattern seen in past embedded-device vulnerability disclosures.

The results: 99 vulnerabilities, 66 zero days, 90 hours

The headline numbers from the report are specific. The approach reduced analysis time from around nine months to 90 hours. In that time the team uncovered 99 vulnerabilities, including 66 zero days. A zero day, in this context, means a previously unknown vulnerability for which no patch existed at the time of discovery.

How the hours actually broke down

The 90 hours was not all machine time. After 10 hours of AI analysis, the team then spent around 80 hours manually validating the findings. That split matters. Roughly nine out of ten hours in this workflow were human expert hours spent confirming that what the AI flagged was real and exploitable. The AI compressed the discovery phase; humans still did the verification.

The cost figures

Forescout also priced the exercise. Static analysis across the 10 stacks cost around $935, which the report calculates as $2.58 per finding. The report adds an important caveat for anyone running proprietary or embedded software: binary analysis was around six times more expensive and took twice as long as source-code analysis. Source code is what you get with open source software. Binaries are what you get with most commercial and embedded products, where the researcher works from compiled code without the original source.

What Forescout itself said about the result

"AI did not eliminate the need for expert vulnerability researchers, but it dramatically compressed the work. 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," said Daniel dos Santos, VP of Research at Forescout.

Dos Santos continued: "Our 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. This is especially impactful for critical infrastructure, where legacy and difficult-to-patch devices may remain in service for years."

How AI Vulnerability Discovery Works, and Why It Matters

Strip away the vendor framing and the mechanism is straightforward. Traditional vulnerability research on embedded software is slow because a skilled researcher must read large volumes of code, form hypotheses about where flaws might live, and manually probe each one. The Forescout harness automated the hypothesis generation and initial triage stages. The AI did the broad, repetitive sweep across the code; the researchers then confirmed which findings were genuine.

Why does this matter to a business reader rather than a researcher? Because the same economics apply to both sides. Forescout's own words make the point explicitly: vulnerabilities in complex embedded software can now be found much faster and at a fraction of the cost, by security researchers and attackers alike. The nine-month window that used to separate a flaw's existence from its discovery is, for this class of target, no longer a reliable buffer. An adversary with modest budget and off-the-shelf models can run a similar workflow against the firmware in a device you operate.

The cost asymmetry deserves attention. Around $935 in static analysis across 10 stacks, at $2.58 per finding, is a rounding error against the potential value of a single exploitable zero day in widely deployed networking code. When discovery gets this cheap, disclosure volume rises, and the window between a vendor's patch release and broad attacker awareness narrows. Forescout's guidance to accelerate patching timelines when fixes exist is a direct consequence of that math.

There is also a structural point for critical infrastructure operators. Legacy and difficult-to-patch devices may remain in service for years, as dos Santos noted. If discovery accelerates but your patching cycle does not, the gap between known risk and remediated risk widens. This is why Forescout paired its findings with recommendations to prioritize by reachability, exploitability, and operational impact, and to isolate devices that cannot be patched, such as those that have reached end of life. Those recommendations align with the direction of federal baseline guidance: CISA's Cross-Sector Cybersecurity Performance Goals 2.0 describe a common set of protections all critical infrastructure entities should implement to reduce known risks, and NIST SP 800-61 Rev. 3 frames incident response as part of ongoing cybersecurity risk management under the Cybersecurity Framework 2.0 functions.

Finally, the workflow itself is a template worth understanding. Ten hours of machine analysis, then 80 hours of human validation. Organizations adopting AI in security operations should expect a similar shape: AI widens the funnel, humans close it. Any claim that AI alone finds and fixes vulnerabilities, without expert validation, does not match what this report actually describes.

What the Evidence Does and Does Not Show

The report is credible as a demonstration, but it is a demonstration of one workflow against one class of target. Read the limits carefully before generalizing.

What is established

Forescout built an AI-assisted vulnerability research harness, ran it against 10 popular TCP/IP stacks, and reported the time, cost, and finding counts above. Those figures come from the company's own report. The team validated findings manually, and the report is explicit that expert researchers remained essential to the workflow.

What is not established

The sources do not state how the 10 stacks were selected, what models powered the harness, how the harness was built, or how the 99 vulnerabilities were disclosed and remediated. The report does not say whether the 66 zero days have since been patched, by which vendors, or on what timelines. It also does not claim the same results would transfer to other software classes, larger codebases, or targets with different architectures. Binary analysis being six times more expensive and twice as slow than source-code analysis is itself a limit: most commercial and embedded products ship as binaries, so the cheapest results apply where source is available.

AI cuts both ways, and not always in the attacker's favor

The Infosecurity coverage is careful on this point. Research by ReliaQuest points to AI helping threat actors make attacks cheaper, faster to scale, easier to customize, and harder to spot. At the same time, some experts have warned AI can make attacks less efficient. One security expert from Proofpoint recently described AI-generated exploits as highly flawed. The honest reading is that AI lowers the cost of offensive work in aggregate while producing uneven quality, and defenders should not assume either worst-case or best-case outcomes.

Open questions for defenders

Several things remain unknown. It is not known how quickly this kind of AI-assisted research will spread beyond well-resourced research teams, or how vendors of embedded devices will respond in their own disclosure and patching practices. It is not known whether the 90-hour figure is repeatable across different target classes or whether it reflects a favorable first run. And the report does not measure how defenders' existing prioritization and patching processes hold up under a faster disclosure cadence. Forescout's recommendations, from risk-based prioritization to reducing exposure of services such as FTP, Telnet, IPv6, and web administration interfaces, are a response to that uncertainty rather than a measured outcome of it.

For organizations in regulated industries, the practical implication is not panic but tempo. The discovery side of the equation has compressed. The response side, prioritization, patching, isolation of unpatchable devices, and procurement requirements such as supplier support lifetimes, vulnerability disclosure policies, and software bills of materials, is where the report points next. We will cover what to actually do about it in Part Two.

The Compliance and Risk Implications for Regulated Organizations

The Forescout research changes the risk calculus that compliance frameworks were built around. Most regulatory expectations for vulnerability management assume a certain pace of discovery and disclosure. When AI-assisted vulnerability discovery compresses analysis from roughly nine months to 90 hours, the window between a flaw existing in the wild and a flaw being known, weaponized, or disclosed shrinks dramatically. That affects how organizations in regulated industries must demonstrate diligence, not just how they patch.

For defense contractors, the obligations are concrete. Whatever the assessment timeline, the underlying requirement to protect covered defense information does not pause. DFARS 252.204-7012 separately requires reporting a cyber incident affecting covered defense information to DoD within 72 hours of discovery, and NIST SP 800-171's Incident Response family requires an operational incident handling capability, not merely an occasional external report. Faster discovery on the attacker side means your detection and response capability has to be genuinely operational, and your evidence of it has to be current.

For healthcare entities, the HIPAA Security Rule at 45 CFR Part 164 requires a Risk Analysis under 45 CFR 164.308(a)(1)(ii)(A). A risk analysis that treats all vulnerabilities as equal, or that relies on an annual scan cycle, will not reflect a threat environment where 66 zero days can be surfaced in a single research effort. The economics described in the Forescout report apply to attackers and researchers alike, which means the population of exploitable flaws in the medical devices, imaging systems, and network equipment healthcare organizations depend on can grow faster than annual assessment cycles anticipate.

For organizations working toward SOC 2, ISO/IEC 27001:2022, or PCI DSS v4.0.1, the implication is less about a specific clause and more about the reasonableness standard that auditors and assessors apply. Demonstrating that you prioritize by reachability, exploitability, and operational impact, and that your patching timelines account for faster disclosure, is increasingly the posture a reasonable organization is expected to hold. CISA's Cross-Sector Cybersecurity Performance Goals 2.0, aligned to the NIST Cybersecurity Framework 2.0 functions including GOVERN, offer a baseline benchmark for exactly this kind of prioritized, high-impact practice (https://www.cisa.gov/cross-sector-cybersecurity-performance-goals). NIST SP 800-61 Rev. 3 similarly frames incident response as something to integrate throughout cybersecurity risk management under CSF 2.0, not as a standalone plan filed on a shelf (https://csrc.nist.gov/pubs/sp/800/61/r3/final).

What This Means for Regulated Industries

Defense Contractors

The Forescout research targeted TCP/IP stacks, the software layer that sits inside nearly every embedded and networked device in a defense industrial base environment. If 99 vulnerabilities, including 66 zero days, can be found across 10 popular stacks for around $935 in static analysis, assume that the firmware in switches, routers, and industrial controllers on your network is being examined by someone other than the vendor. Three actions follow. First, know your exposure: maintain an accurate inventory of network-facing devices and reduce unnecessary exposure by disabling services such as FTP, Telnet, IPv6, and web administration interfaces, as Forescout recommends. Second, accelerate patching when fixes exist, because disclosure rates are rising. Third, isolate devices that cannot be patched, including anything at end of life, rather than accepting their risk by default. If your environment holds CUI, these steps also feed directly into the system security plan and POA&M evidence a C3PAO or your own self-assessment will examine.

Healthcare

Clinical environments are full of the exact class of device this research describes: legacy, difficult to patch, and expected to remain in service for years. Forescout's own VP of Research, Daniel dos Santos, made that point explicitly about critical infrastructure. For healthcare leaders, the practical translation is segmentation and reachability. A vulnerability in an unpatchable imaging system matters far less if that system cannot reach the internet or the electronic health record network, and far more if it can. Pair that with a risk analysis that documents why each legacy device is accepted, isolated, or scheduled for replacement. Given that IBM's 2026 research puts the average healthcare breach at $6.64M, and HHS filings show business associates accounted for about 66% of breached healthcare records in 2024, third-party and device-level exposure deserve the same scrutiny as your own systems.

Legal

Law firms sit at the intersection of client confidentiality obligations and an increasingly exposed software supply chain. The practical risk here is not that your firm runs TCP/IP stack research; it is that the vendors you rely on, from practice management platforms to e-discovery providers, run embedded software with the same flaw density Forescout found. Forescout's procurement guidance is directly applicable: require suppliers to define support lifetimes, publish vulnerability disclosure policies, provide software bills of materials (SBOMs), and demonstrate ongoing security testing. Make those requirements part of vendor onboarding, not an annual questionnaire. Firms handling matters for defense contractors or healthcare clients should also expect those clients to flow down security expectations, and should be able to answer questions about patch cadence and incident response with evidence rather than assurances.

Financial Services

Financial institutions already operate under continuous examination, and the Forescout findings reinforce a trend examiners have been signaling: point-in-time vulnerability scans are losing their value as a control. The report's core lesson, that prioritization should be based on reachability, exploitability, and operational impact rather than treating all vulnerabilities equally, maps well onto how a mature financial institution should already be running its vulnerability management program. The new pressure is speed. If fixes exist, patching timelines need to compress to account for a faster rate of vulnerability disclosure. Institutions relying on third-party fintech vendors and core processing providers should apply the same procurement requirements Forescout recommends, including SBOMs and demonstrated ongoing security testing, and should verify that incident response coordination meets the expectations of NIST SP 800-61 Rev. 3's integrated approach (https://csrc.nist.gov/pubs/sp/800/61/r3/final).

An Action Plan for This Month

You do not need to replicate Forescout's research harness to act on its findings. These steps can be started within weeks by most security and IT leaders.

  1. Inventory your network-facing devices. Identify every device running an embedded TCP/IP stack, including switches, routers, firewalls, controllers, and any vendor appliance. You cannot prioritize what you have not enumerated.
  2. Disable unnecessary services. Following Forescout's guidance, turn off FTP, Telnet, IPv6, and web administration interfaces wherever they are not required. This is one of the fastest exposure reductions available and costs nothing.
  3. Identify unpatchable and end-of-life devices. Build a list of devices that cannot receive fixes, and begin isolating them on segmented network zones rather than leaving them flat on the production network.
  4. Re-baseline your patching SLAs. Compare your current remediation timelines against a faster disclosure environment. Where fixes exist, shorten the window between vendor release and deployment, prioritizing internet-reachable systems.
  5. Shift prioritization to reachability and exploitability. Stop working vulnerability queues in raw severity order. Rank by whether the flaw is reachable, exploitable, and operationally impactful, exactly as Forescout recommends.
  6. Update your incident response capability against NIST SP 800-61 Rev. 3. Confirm you have an operational incident handling capability, not just a document, and that detection, response, and recovery are integrated with your risk management activities (https://csrc.nist.gov/pubs/sp/800/61/r3/final).
  7. Insert security requirements into procurement. Add supplier requirements for defined support lifetimes, published vulnerability disclosure policies, SBOMs, and demonstrated ongoing security testing to your vendor onboarding process now, before your next major purchase.
  8. Verify your reporting readiness. If you are a defense contractor, confirm you can meet the DFARS 252.204-7012 requirement to report covered cyber incidents to DoD within 72 hours of discovery, including having the required medium assurance certificate obtained in advance and the 90-day evidence preservation process ready.
  9. Benchmark against CISA's CPGs 2.0. Use the Cross-Sector Cybersecurity Performance Goals 2.0 as a gap-check on your essential practices, particularly for IT and OT environments where legacy devices persist (https://www.cisa.gov/cross-sector-cybersecurity-performance-goals).
  10. Get an outside read on your exposure.

How Petronella Technology Group, Inc. Helps

The Forescout report describes a world where vulnerability discovery is fast, cheap, and available to everyone. Responding to that world requires three capabilities: continuous visibility into what is exposed, disciplined prioritization and patching, and response capacity that is genuinely operational. Petronella Technology Group, Inc. provides all three, delivered remote-first across all 50 states.

Our Managed Detection and Response service combines 24/7 threat monitoring, AI-powered analytics, and human-led investigation, replacing the need for an in-house security operations center. Our AI-augmented hybrid SOC runs ten-plus production AI agents on the company's private AI cluster, with guardrails that matter in regulated environments: the AI never closes a ticket on its own, never touches production systems without human authorization, and every action is logged for CMMC and HIPAA audit. Human analysts hold CMMC-RP credentials, and a Digital Forensics Examiner is on staff. MDR reporting is designed to provide the evidence auditors need for HIPAA, CMMC, PCI DSS, SOC 2, and NIST 800-171 verification, and full coverage with tuned detection is operational within 30 days of contract signing.

For organizations that need broader program support, our cybersecurity services deploy a 39+ layer security stack covering People, Process, and Technology, including MFA, FIPS-validated encryption, audit logging, and configuration management baselines. Because the stack runs on our own private AI cluster, defense industrial base and healthcare clients that cannot send CUI or PHI to a public-cloud SOC are fully covered.

On the compliance side, our compliance practice covers CMMC, HIPAA, SOC 2, ISO/IEC 27001:2022, and PCI DSS, with a vCISO offering that delivers a written current-state assessment within 30 days and a 90 to 180 day remediation roadmap written for the executive reader. Every practitioner on the compliance team holds the CMMC-RP designation, and the firm is a Cyber AB Registered Provider Organization, RPO #1449. If your organization is also deploying AI internally, our analysis of Shadow AI: Discover and Secure Enterprise LLMs addresses the governance side of the same equation.

The through-line is simple: attackers and researchers now find vulnerabilities at the same accelerated pace. The organizations that fare best are those with continuous detection, fast and risk-based remediation, and audit-ready evidence. That is what we build.

Sources

Frequently Asked Questions

How did AI reduce vulnerability discovery from nine months to 90 hours?

Forescout Technologies used a specially built AI-assisted vulnerability research harness, running publicly available models, to analyze 10 popular TCP/IP stacks. The AI performed about 10 hours of analysis, after which the team spent roughly 80 hours manually validating findings, for a total of about 90 hours compared with around nine months previously (https://www.infosecurity-magazine.com/news/ai-cut-vulnerability-discover-nine/).

What did the Forescout AI vulnerability research actually find?

The research uncovered 99 vulnerabilities across the 10 stacks, including 66 zero days. Static analysis across all 10 stacks cost around $935, or $2.58 per finding, though binary analysis was around six times more expensive and took twice as long as source-code analysis (https://www.infosecurity-magazine.com/news/ai-cut-vulnerability-discover-nine/).

Does this mean attackers can now find zero days cheaply?

Yes, that is the core warning. Daniel dos Santos, Forescout's VP of Research, stated that with 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. Related research from ReliaQuest indicates AI helps threat actors make attacks cheaper, faster to scale, easier to customize, and harder to spot, although a Proofpoint expert has noted AI-generated exploits can be highly flawed (https://www.infosecurity-magazine.com/news/ai-cut-vulnerability-discover-nine/).

What should organizations do about unpatchable embedded devices?

Forescout recommends isolating devices that cannot be patched, such as those that have reached end of life, and reducing unnecessary exposure by disabling services such as FTP, Telnet, IPv6, and web administration interfaces. Prioritization should be based on reachability, exploitability, and operational impact rather than treating all vulnerabilities equally (https://www.infosecurity-magazine.com/news/ai-cut-vulnerability-discover-nine/).

Does faster vulnerability discovery change my compliance obligations?

The obligations themselves have not changed, but the diligence expected to meet them has. NIST SP 800-61 Rev. 3 calls for incident response integrated throughout cybersecurity risk management under the NIST Cybersecurity Framework 2.0 (https://csrc.nist.gov/pubs/sp/800/61/r3/final), and CISA's Cross-Sector Cybersecurity Performance Goals 2.0 provide a baseline benchmark for critical infrastructure operators to measure and improve maturity (https://www.cisa.gov/cross-sector-cybersecurity-performance-goals). Faster disclosure rates make annual, scan-driven approaches harder to defend as reasonable.

How can a small security team keep up with AI-accelerated threats?

Focus on the high-impact fundamentals: accurate inventories, reduced exposure, segmentation of unpatchable devices, risk-based prioritization, and 24/7 detection and response. For organizations without the staffing for an internal SOC, outsourced options such as Managed Detection and Response provide continuous monitoring and human-led investigation without the cost of building one. Start with a conversation through our contact form or the number below.

If this article raised questions about your organization's exposure, call Penny, our AI assistant, at 919-348-4912 to book a consultation with Petronella Technology Group, Inc., or reach us through our contact form.

Get the 2026 Cybersecurity Survival Guide

Free, practical, and specific to regulated environments. We will email it to you.

No spam. Unsubscribe anytime.

Talk to Petronella Technology Group, Inc.
Private, on-premises AI and compliance for regulated data. Call 919-348-4912, get a free AI assessment, or explore our AI, cybersecurity, and compliance services.