We Built a Vulnerability Vending Machine: AI Tokens in, Zero-Days Out
Researchers at Intruder used AI tokens to find a zero-day vulnerability in a popular WordPress plugin, demonstrating the potential of AI in vulnerability research.
Intelligence analysis by Llama
Intruder's team used a pipeline that combined code scanning with AI to find a SQL injection vulnerability in the Creative Mail plugin, which has over 300,000 users. The vulnerability allows an attacker to read the database, including admin hashes and secret tokens.
Imagine you have a big box of LEGOs, and you want to find a specific LEGO piece. A traditional code scanner would be like looking through the whole box at once, which is hard and time-consuming. But a new way of using AI is like using a special tool that can look at one small part of the box at a time, and it can find the LEGO piece you're looking for much faster. This is what Intruder did to find a vulnerability in a popular WordPress plugin.
Analysis
The Focus Problem: Why Pointing AI at a Whole Codebase Doesn't Work
LLMs are excellent at taking small segments of code, or a description of a specific problem, and finding an interesting solution. However, when pointed at a large codebase and asked to find security issues, it will try to ingest every file in the repo. This is expensive in tokens and worse for accuracy, as the model's context becomes full of irrelevant code, and the bug you actually want is buried in noise.
Traditional code scanning frameworks already solve this problem. Intruder uses a technique called a program slice, which is similar to when an IDE or LSP tool uses features like 'find implementation' or a call graph to find all functions called by the current function. These are mature, well-tested tools, and they sidestep the diluted-context problem entirely.
Our Pipeline: From Codebase to Working Exploit
Intruder built a pipeline that takes a codebase, runs it through a code scanning engine (Joern), generates slices of code relevant to each finding, and uses an LLM to triage and exploit the issue. The design was inspired by nooperator's work on Slice, though they use Joern rather than CodeQL and designed the slicing algorithm quite differently to handle the specific vulnerability classes they're looking for.
The First Vulnerability: A Blind SQL Injection in a Popular WordPress Plugin
The first bug the pipeline vended was CVE-2026-3985, a SQL injection vulnerability in the Creative Mail plugin. It stood out to them for a few reasons: It's high impact, giving an attacker read access to the database (including admin hashes and secret tokens!). It requires multiple chained requests to exploit, making it less likely to be detected by traditional tooling. The root cause was hidden from the developer's own static analysis tooling by a mistake in their code.
Exploitation does require WooCommerce to be installed alongside Creative Mail, but since WooCommerce is a common reason people run WordPress (over 7 million active installs), the combination is common. The exploitation agent one-shotted a working proof-of-concept, producing a check to confirm the issue existed and a full extraction method capable of pulling password hashes from the database.
Key points
- Intruder used a pipeline that combined code scanning with AI to find a zero-day vulnerability in a popular WordPress plugin.
- The vulnerability allows an attacker to read the database, including admin hashes and secret tokens.
- The pipeline used a technique called a program slice to focus on relevant code and avoid noise.
- The exploitation agent one-shotted a working proof-of-concept, producing a check to confirm the issue existed and a full extraction method capable of pulling password hashes from the database.
This development could lead to more efficient and effective vulnerability research, allowing security teams to identify and fix issues before they can be exploited. It could also lead to the discovery of more complex and hard-to-find vulnerabilities, which would improve the overall security of software.
The use of AI in vulnerability research could also lead to the creation of more sophisticated and targeted attacks, which could be difficult to detect and mitigate. Additionally, the reliance on AI could lead to a decrease in human security expertise, as machines take over some of the tasks currently performed by humans.
