Finite State to Reveal Results of Six-Month, AI-Driven Firmware Security Research at Hardwear.io 2026

Finite State, a leader in product security and software supply chain risk management, today announced that Roland Lindsey, Chief Product Security Engineer, will present findings from six months of intensive firmware security research at Hardwear.io Netherlands 2026, taking place November 19–20 at the Amsterdam Marriott Hotel.

The 90-minute technical session, “140+ Detonations: What Six Months of LLM-Driven Firmware Teardowns Actually Proved,” will examine how large language models (LLMs), automated code analysis, adversarial validation, and dynamic testing can be combined to identify and verify exploitable vulnerabilities in connected devices.

Over six months, Lindsey and the research team analyzed 35 firmware images spanning embedded Linux routers, industrial controllers, automotive electronic control units (ECUs), and bare-metal real-time operating system (RTOS) devices. Across the broader research team, more than 50 firmware images were examined.

Using an LLM-driven analysis pipeline, the research identified 802 potential remote code execution (RCE) vulnerabilities involving command and code injection that survived initial triage. However, only 68 were subsequently demonstrated through reproducible proofs of vulnerability against the shipped binaries. Including successfully exploited memory-corruption vulnerabilities, the total number of demonstrated exploits exceeded 140.

The findings underscore a fundamental challenge facing manufacturers and security researchers: Identifying a potentially exploitable vulnerability is not the same as proving that an attacker can exploit it.

As AI makes automated vulnerability discovery faster and more accessible, distinguishing actionable security risks from false positives and unproven findings is becoming increasingly important.

Lindsey’s presentation will examine the research methodology, including automated firmware unpacking, decompilation, parallel AI-assisted vulnerability hunting, adversarial validation, and dynamic exploit verification under emulation.

Key takeaways:

A reusable teardown pipeline you can run yourself — unpack → decompile → fan-out hunt → adversarial triage → dynamic proof — not a vendor black box.

Why proving beats flagging, and how to tell them apart — the discipline that separates “reachable, sink confirmed” from a detonated root shell, and the guards that stop a model over-claiming its own exploit.

The negative results, so you skip the dead ends — what we were sure would help (e.g. reachability hints) and measurably didn’t.

An honest cost/quality picture — where the false positives actually live, what a teardown costs, and the bug classes automation still misses.

The session will also address unexpected findings and limitations, including:

Why verification matters: The substantial difference between vulnerabilities flagged during analysis and those demonstrated through reproducible exploitation.

The value of adversarial AI validation: How challenging and attempting to disprove initial findings can improve the reliability of automated vulnerability discovery.

Unexpected limitations: Why techniques such as reachability hints, initially expected to improve results, failed to provide measurable benefits in this research.

Practical barriers to exploitation: The challenges posed by emulation environments, complex execution conditions, and hardware architectures lacking established testing tools.

Real-world costs and limitations: The resources required for firmware analysis and the vulnerability classes that automated techniques still fail to detect.

Attendees will gain a reproducible methodology for conducting AI-assisted firmware security assessments, along with practical guidance for distinguishing credible vulnerability findings from demonstrated security risks.

Presentation Details

Session: 140+ Detonations: What Six Months of LLM-Driven Firmware Teardowns Actually Proved
Speaker: Roland Lindsey, Chief Product Security Engineer, Finite State
Date: Thursday, November 19, 2026
Time: 1:30–3:00 p.m. CET
Location: Amsterdam Marriott Hotel, Amsterdam, Netherlands
Event: Hardwear.io Netherlands 2026
Event website: https://hardwear.io/nl-2026/

About Finite State

Finite State empowers device OEMs to ship securely while enabling engineering teams to move at the speed of AI, immediately transforming product artifacts into audit-ready assurance through a single automated workflow. Leveraging deep binary analysis and AI-native execution, the platform unifies code, compiled components, and firmware in minutes – connecting security design with deployed software. By continuously generating SBOMs, VEX, and signed compliance packages, Finite State enables connected device companies across industries such as medical devices and automotive to meet evolving regulations, including the EU Cyber Resilience Act (CRA), and deliver continuous compliance at speed.

Learn more at https://finitestate.io/

Media gallery