# Forescout: an AI harness found 99 TCP/IP stack bugs in 90 hours

> Forescout reports an AI-assisted pipeline found 66 unknown bugs in embedded TCP/IP stacks for $1,200 to $2,300 in model spend.

- **Severity:** Info: Research, tools, evals or lab claims with no direct exposure. Context, not an alert.
- **Category:** Research
- **Published by:** Forescout
- **Disclosed:** 2026-11-12
- **Affects:** Contiki, FNET, PicoTCP, Nucleus NET, wolfIP, lwIP, Ethernut
- **Research by:** Stanislav Dashevskyi (Forescout Vedere Labs), Francesco La Spina (Forescout Vedere Labs), Daniel dos Santos (Forescout Vedere Labs)
- **Primary source:** https://www.forescout.com/research-labs/finding-critical-infrastructure-bugs-with-off-the-shelf-ai/
- **Page:** https://theinjection.dev/items/forescout-off-the-shelf-ai-tcp-ip-bugs/

Forescout Research says an AI harness built on Claude models found 99 vulnerabilities in 10 embedded TCP/IP stacks in about 90 hours, 66 of them previously unknown. The team estimates the same work would have taken nine months by hand.

## What happened

Forescout Research published Finding Critical Infrastructure Bugs with Off-the-Shelf AI on November 12, 2026. Stanislav Dashevskyi, Francesco La Spina and Daniel dos Santos applied an AI-assisted pipeline to 10 embedded TCP/IP stacks, seven of them revisited from the team's earlier Project Memoria work. The run confirmed 99 vulnerabilities, 66 of which were previously unknown.

## How it works

The pipeline chains ordinary static analysis with model-driven triage rather than replacing either. Semgrep and Joern, running 32 custom rules, narrow the code to candidate sites, and Ghidra handles the closed-source stacks. Claude Sonnet 4.6 drives the first three stages at high and max reasoning, and Opus models between 4.6 and 5 verify candidates dynamically. Forescout reports precision between 11% and 70% depending on the stack, so human validation still consumed most of the 90 hours.

## Who is affected

The flaws sit in embedded networking code, not in AI products: Contiki, FNET, PicoTCP, Nucleus NET, wolfIP, lwIP and Ethernut, plus three closed-source stacks Forescout labels S1 to S3. Twenty-nine findings in those proprietary stacks were still under vendor investigation at publication. Forescout reports that seven flaws sat in a stack used by an unsupported wireless access point, including hardcoded administrator credentials.

## What to do

Prioritize embedded device risk by reachability, exploitability and operational impact, and restrict exposed services such as FTP, Telnet, IPv6 and web admin interfaces. Isolate devices that cannot be patched, and build support lifetimes, disclosure policies and SBOMs into procurement. Forescout advises maintainers to rerun variant analysis over previously audited code and to expand triage capacity for a higher volume of AI-generated reports.

## Quick facts

- **Stacks tested:** 10 embedded TCP/IP stacks, seven revisited from Project Memoria and three new
- **Findings:** 99 confirmed, 66 previously unknown and 33 rediscovered
- **Severity:** Forescout rates 52% of the 66 new findings critical or high
- **Reported:** 54 of the 66 new findings, 20 open source and 34 proprietary
- **Fixed:** 24, across Contiki, FNET, PicoTCP, Nucleus NET, wolfIP and lwIP
- **Model spend:** $935 for static analysis, $1,200 to $2,300 estimated in total
- **CVEs:** Requested for the open source findings, not yet assigned

## Questions

### Does the Forescout research mean AI can find vulnerabilities without expert researchers?

Forescout's report argues the opposite. Daniel dos Santos states that AI did not remove the need for expert vulnerability researchers but dramatically compressed the work. Roughly 10 of the 90 hours were AI analysis, and the remaining time was human validation. Reported precision ranged from 11% to 70% per stack, so most candidate findings still needed a researcher to confirm or reject them.

### Have CVE ids been assigned to the vulnerabilities Forescout found?

Forescout requested CVE ids for the open source findings, but the report states that none had been assigned when it was published. Twenty-four of the reported flaws are already fixed, across Contiki, FNET, PicoTCP, Nucleus NET, wolfIP and lwIP. Twenty-nine findings in three closed-source stacks remained under vendor investigation, and the Ethernut maintainer had not responded to the disclosure.

### Which devices run the TCP/IP stacks Forescout tested?

Embedded and critical infrastructure devices run these stacks, which Forescout tested as Contiki, FNET, PicoTCP, Nucleus NET, lwIP, wolfIP, Ethernut and three closed-source stacks. Forescout highlights legacy exposure: seven flaws sat in a stack used by an unsupported wireless access point, including hardcoded administrator credentials, which the vendor may never patch. Asset owners should isolate devices that cannot be updated.

## Sources

1. [Forescout Research](https://www.forescout.com/research-labs/finding-critical-infrastructure-bugs-with-off-the-shelf-ai/)
2. [Forescout Announcement](https://www.financialcontent.com/article/bizwire-2026-11-12-forescout-research-shows-ai-cut-vulnerability-hunting-from-nine-months-to-90-hours)
3. [Cybersecurity Dive](https://www.cybersecuritydive.com/news/ai-vulnerabilities-critical-infrastructure-embedded-software/874499/)