← Voltar ao feed

Forescout Vedere Labs uses AI to find 99 flaws, 66 new, in ten legacy TCP/IP stacks used in OT, IoT and critical infrastructure

5 fonte(s) · 5 atualização(ões) · publicado 12/11/2026 12:30 · coletado há 1 dia

Matéria consolidada. Reúne 6 matérias que o Tifon tinha publicado separadamente sobre a mesma história. A IA reescreveu o relatório com o que as 5 fontes trouxeram em 16/11/2026 12:09; os links antigos trazem para cá.

Fontes: digit.fyi ↗ infosecurity-magazine.com ↗ helpnetsecurity.com ↗ industrialcyber.co ↗ cybersecuritydive.com ↗

Forescout Research's Vedere Labs used AI-assisted vulnerability analysis on legacy TCP/IP stacks and found 99 vulnerabilities across ten widely used stacks, 66 of them previously unknown (industrialcyber.co, 2026-11-13). The work is described in the report Finding Critical Infrastructure Bugs with Off-the-Shelf AI (infosecurity-magazine.com). The team built a harness around Claude models and existing security tools (helpnetsecurity.com). It cut research time from about nine months to 90 hours, roughly 10 hours of AI analysis plus 80 hours of human validation. 52% of the new flaws are rated high or critical (helpnetsecurity.com, digit.fyi). The stacks underpin products in operational technology (OT), IoT, medical devices and critical infrastructure, so the findings are a supply chain risk. Siemens has identified affected building automation controllers in its APOGEE, Desigo and TALON lines, and other vendors are still investigating (helpnetsecurity.com). Forescout reported 54 of the new flaws to vendors and maintainers, and several projects have issued fixes (helpnetsecurity.com). Cybersecuritydive.com (2026-11-12) first covered the research more generally. It described AI finding widespread IoT vulnerabilities in critical infrastructure embedded software faster and more cheaply than human-led methods. The sources name no CVE identifiers and none reports exploitation in the wild or threat actors. The stack names are not given, and the names of three stacks are withheld while investigations stay open.

Entenda a falha

A vulnerabilidade

No CVE IDs are listed. Forescout's Vedere Labs confirmed 99 vulnerabilities across 10 TCP/IP stacks, the networking code in OT, IoT, medical and network devices. 66 were previously unknown, and 52% of those are rated high or critical (47% high, 5% critical, 45% medium, 3% low). Most stacks are not named, and three are withheld during disclosure. By type, memory corruption is 49%, integer errors 27%, input validation 16% and logic bugs 8%. Siemens has found affected APOGEE, Desigo and TALON controllers, and other vendors are still investigating. No affected or fixed versions are given. A discontinued access point also had seven device-specific flaws, including unchangeable administrator credentials and a buffer overflow in firmware upload.

Como funciona

The affected code is the part of a TCP/IP stack that receives and parses data arriving over the network, in the core IP, TCP, UDP, ICMP and Ethernet handling. Network traffic reaches the stack, the stack parses it, and the failure happens where lengths, bounds or values are not handled correctly. The result is out-of-bounds reads and writes, buffer overflows, integer overflow or truncation, infinite loops or null dereferences. In the access point, the overflow happens in the firmware upload process. The articles do not give the exact sequence for each stack.

Por que é vulnerável

The flaws are mostly missing or wrong checks in protocol-parsing code: bounds, integer handling, null-termination and input validation. Logic and design flaws also appear, such as PRNG issues, but the tool found those hard to detect. The access point has hardcoded administrator credentials that cannot be changed. These bugs survived partly because manual review is slow: Project Memoria took months of manual analysis and missed some of them. Fixes must travel from the stack developer through firmware and product vendors, and discontinued products may never get one. The articles do not describe what each fix changes.

Como foi descoberta

Forescout researchers built an AI harness around Claude models and existing 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 assessed reachability and exploitability, and a final stage ran the software and built triggering tests under researcher oversight. It produced 362 findings, 312 after deduplication, and 99 were confirmed. It rediscovered 33 of 82 known Project Memoria+ flaws. Analysis took about 10 hours plus 80 hours of validation. Forescout reported 54 new flaws (20 open source, 34 proprietary) to vendors and maintainers. Industrial Cyber says only five were fixed so far, while Help Net Security says several projects have issued fixes. No exploitation in the wild is reported.

PoC existente, privada

O exploit existe, mas quem o tem não o publicou.

No exploit was published. The pipeline's final stage runs the software and builds a test that triggers each issue, so Forescout holds verification-level triggering tests for the confirmed flaws, and none were released. Forescout warns that attackers can use the same low-cost workflow, so patching or isolating exposed and legacy devices is urgent.

Explicação gerada por IA a partir da matéria e dos artigos, em 16/11/2026 12:24.

Detalhes técnicos

  • The new flaws are mainly memory corruption, integer errors and input validation issues (industrialcyber.co).
  • The affected stacks are embedded in many vendors' products, so one flaw can reach many devices (industrialcyber.co). Based on earlier Vedere Labs work, flaws in these components are likely to affect hundreds of downstream products from dozens of vendors across OT, IoT, medical and network equipment (digit.fyi).
  • Legacy devices are often unsupported and hard to patch, so vulnerabilities can stay unmitigated for years (industrialcyber.co).
  • Cybersecuritydive.com says the persistent weaknesses in critical infrastructure embedded software could be exploited if not addressed promptly. It also says AI-assisted assessment is faster and cheaper than traditional human-led research.
  • The harness was applied to 10 TCP/IP stacks: seven that Vedere Labs had analysed before and three new ones (digit.fyi).
  • Pipeline: code scanners using 32 rules derived from Project Memoria, then AI agents assessing reachability and exploitability, then a final stage that runs the software and builds triggering tests under researcher oversight (helpnetsecurity.com).
  • Automated analysis took about 10 hours, followed by about 80 hours of validation. Project Memoria took roughly nine months of manual code analysis (helpnetsecurity.com).
  • Automated code inspection cost $935, or $2.58 per finding. Total model costs ranged from about $1,200 to $2,300 (helpnetsecurity.com, infosecurity-magazine.com, digit.fyi).
  • The run produced 362 initial findings and 312 candidates after deduplication, and 99 were confirmed. The system rediscovered 33 of 82 known vulnerabilities, about 40% (helpnetsecurity.com, digit.fyi).
  • 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 (helpnetsecurity.com).
  • Compiled targets without source code cost about six times as much and took about twice as long, which shows the approach also works on proprietary firmware (helpnetsecurity.com, infosecurity-magazine.com, digit.fyi).
  • 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 (helpnetsecurity.com).
  • Forescout notes that attackers can use the same low-cost workflow to find flaws in embedded and critical-infrastructure software (infosecurity-magazine.com, helpnetsecurity.com).

Remediação

Forescout's guidance, as reported by industrialcyber.co:

  • Prioritize vulnerabilities by exploitability and operational impact.
  • Minimize device exposure.
  • Isolate devices that cannot be patched.
  • Accelerate patch deployment.
  • Build security requirements into procurement.
  • Vendors should integrate AI-assisted testing into their development pipelines to detect vulnerabilities proactively.

Additional guidance:

  • Prioritise risk by reachability, exploitability and operational impact. Disable unnecessary services such as FTP, Telnet, IPv6 and web administration interfaces (infosecurity-magazine.com).
  • Require suppliers to define support lifetimes, publish vulnerability disclosure policies, provide SBOMs and demonstrate ongoing security testing (infosecurity-magazine.com).
  • For devices that cannot be patched, Forescout's CEO says the priority should be containing risk rather than patching (helpnetsecurity.com). Forescout advises asset visibility, risk-based prioritisation and containment controls (digit.fyi).
  • Fixes must pass from the software developer through firmware and product vendors, and discontinued products may never get a patch (digit.fyi). Several projects have issued fixes for the 54 reported flaws, and disclosure is still ongoing for some (helpnetsecurity.com, digit.fyi).
  • Cybersecuritydive.com adds a call for more attention to AI-assisted security assessments and better defenses for IoT components in critical systems.

Linha do tempo

  1. digit.fyi ↗ · pub. 16/11/2026 10:07 · coletado 16/11 12:00

    Forescout: Off-the-Shelf AI Found 99 Bugs in 10 TCP/IP Stacks, 66 New, in 90 Hours for $935

    ler a análise desta fonte

    Forescout Research – Vedere Labs published "Finding Critical Infrastructure Bugs with Off-the-Shelf AI". The report tests whether generally available AI models and a limited budget can find vulnerabilities in open-source software and proprietary binaries. A purpose-built AI-assisted harness was applied to 10 TCP/IP stacks: seven that Vedere Labs had analysed before and three new ones. It validated 99 vulnerabilities, 66 of them previously unknown, and 52% of the new ones were rated high or critical. Analysis time fell from about nine months to about 90 hours, and the AI rediscovered 40% of known flaws. The article names no CVEs, exploitation in the wild or threat actors, so this is a research and trend finding rather than an active threat.

    Technical Details

    • The 90 hours were about 10 hours of independent AI analysis plus about 80 hours of human validation.
    • Static analysis of the source code cost about $935, or $2.58 per finding.
    • Binary analysis was about six times more expensive and took twice as long as source analysis. It shows the approach also works on proprietary firmware without source code.
    • 54 of the new vulnerabilities were reported to vendors and maintainers. Disclosure is still ongoing for some.
    • Based on earlier Vedere Labs work, flaws in these components are likely to affect hundreds of downstream products from dozens of vendors across OT, IoT, medical and network equipment.

    Remediation

    Forescout argues that containing risk matters more than patching, because legacy devices often cannot be patched or taken offline. It advises asset visibility, risk-based prioritisation of vulnerabilities and containment controls. Fixes must pass from the software developer through firmware and product vendors, and discontinued products may never get a patch. The article does not name the affected stacks or any advisories, and no primary Forescout report was consulted.

  2. infosecurity-magazine.com ↗ · pub. 16/11/2026 06:00 · coletado 16/11 07:01

    Forescout: off-the-shelf AI cut TCP/IP stack vulnerability research from nine months to 90 hours, yielding 99 bugs incl. 66 zero-days

    ler a análise desta fonte

    Forescout published a report, Finding Critical Infrastructure Bugs with Off-the-Shelf AI, on how publicly available AI models and a purpose-built AI-assisted harness sped up vulnerability research in open source software and proprietary binaries. The team analysed 10 TCP/IP stacks and found 99 vulnerabilities, 66 of them zero-days. Analysis time fell from about nine months to 90 hours. This article is a news report on the vendor's research; no CVE IDs, affected-product lists or advisories are named in it, and no web research was done.

    Technical Details

    • About 10 hours of AI analysis was followed by roughly 80 hours of manual validation by experts.
    • Static analysis across the 10 stacks cost about $935, or $2.58 per finding.
    • Binary analysis cost about six times more and took twice as long as source-code analysis.
    • Forescout notes that attackers can use the same workflow, which lowers the cost of finding flaws in embedded and critical-infrastructure software.

    Remediation

    Forescout recommends:

    • Prioritise risk by reachability, exploitability and operational impact.
    • Disable unnecessary services such as FTP, Telnet, IPv6 and web administration interfaces.
    • Isolate devices that cannot be patched, such as end-of-life devices.
    • Speed up patching.
    • Require suppliers to define support lifetimes, publish vulnerability disclosure policies, provide SBOMs and demonstrate ongoing security testing.
  3. helpnetsecurity.com ↗ · pub. 16/11/2026 02:00 · coletado 16/11 04:02

    Forescout uses Claude-based AI harness to find 99 flaws across 10 TCP/IP stacks, including unpatched discontinued devices

    ler a análise desta fonte

    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.
  4. industrialcyber.co ↗ · pub. 13/11/2026 09:36 · coletado 13/11 10:02

    AI-assisted research uncovers 66 new vulnerabilities in legacy TCP/IP stacks affecting critical infrastructure and OT devices

    ler a análise desta fonte

    Forescout Research's Vedere Labs applied AI-assisted vulnerability analysis to legacy TCP/IP stacks, reducing research time from nine months to 90 hours and uncovering 99 vulnerabilities across ten widely used stacks, including 66 previously unknown flaws. Over half of the new vulnerabilities are high or critical severity, primarily involving memory corruption, integer errors, and input validation issues. These stacks underpin products across operational technology (OT), IoT, medical devices, and critical infrastructure, making widespread exploitation a significant supply chain risk. Legacy devices, often unsupported and hard to patch, pose heightened risks as vulnerabilities remain unmitigated for years. Forescout urges prioritizing vulnerabilities by exploitability and operational impact, minimizing device exposure, isolating unpatchable devices, accelerating patch deployment, and incorporating security requirements into procurement. Vendors are encouraged to integrate AI-assisted testing in development pipelines for proactive vulnerability detection. This research exemplifies AI's impact on accelerating and lowering costs of vulnerability discovery, underscoring a pressing need for improved risk management and patching strategies in critical sectors.

  5. cybersecuritydive.com ↗ · pub. 12/11/2026 12:30 · coletado 12/11 13:02 primeiro registro só resumo do feed

    AI accelerates discovery of widespread IoT vulnerabilities in critical infrastructure embedded software

    ler a análise desta fonte

    A recent report highlights how artificial intelligence (AI) has been leveraged to identify widespread vulnerabilities within IoT devices embedded in critical infrastructure software. This demonstrates the increasing effectiveness of AI in accelerating vulnerability research, uncovering potential security flaws more quickly and cost-effectively than traditional human-led methods. The findings underscore persistent weaknesses in critical infrastructure's embedded software that could be exploited if not addressed promptly. This development calls for heightened attention to AI-assisted security assessments and improved defenses for IoT components in critical systems.