Has AI Solved Reverse Engineering?

AI-Assisted Firmware Reverse Engineering

A vulnerability researcher demonstrates how AI accelerated the reverse engineering of firmware from a Dell S3422 monitor. Extracting and decrypting the firmware would traditionally take hours or days, but an AI agent completed the key steps in about 15–16 minutes.

Extracting the Firmware

The monitor’s update is distributed as a Windows .exe file. The researcher uses Binwalk to recursively extract its contents, including the firmware image and Dell’s ISP library, which handles uploading and downloading firmware.

Finding the Decryption Method

Binwalk’s entropy analysis suggests that the firmware image is encrypted. The researcher inspects the ISP library in Ghidra. The AI agent identifies functions that construct a key blob and decrypt the image, extracts the relevant key material, and writes a Python decryption script.

Checking the Result

The agent checks the decrypted image using several indicators:

• A matching MD5 check
• Plausible firmware data and structure
• A valid reset vector
• Content consistent with the monitor’s microcontroller

These checks support the conclusion that the image was correctly decrypted. However, the video does not demonstrate finding or exploiting a monitor vulnerability.

Security Implications

The researcher argues that including the decryption logic and key material in the user-accessible update package undermines the encryption if its purpose is to protect Dell’s intellectual property. A more robust design would keep the decryption key inside the device in non-extractable storage.

More broadly, the demonstration illustrates how AI may reduce the time and expertise needed for reverse engineering, potentially leading to more bugs being found.

Sponsored Segment: AI-Assisted Bug Triage

The video also demonstrates Sentry’s telemetry and MCP integration. An AI agent combines error data with a local codebase, identifies a URL-handling issue, and suggests a fix. The host notes that a growing volume of AI-generated code could increase security debt if bugs are left unaddressed.

Format and Sources

• Reviews, guides, or tutorials: No product review or step-by-step tutorial is presented. The host says the main purpose is to illustrate AI’s effect on reverse engineering.
• Main speakers and sources: The Low Level host and vulnerability researcher narrate the demonstration. The reverse-engineering assistance comes from an AI agent identified in the subtitles as GLM 5.3, using Binwalk and Ghidra. Sentry is featured in a sponsored segment.