Researchers from Google’s OSS-Fuzz team have successfully used AI to identify 26 vulnerabilities in open-source projects maintainers.
These included a flaw that has existed for two decades in the OpenSSL library (CVE-2024-9143), a software library that most HTTPS websites rely on.
The OSS-Fuzz team has supported open-source maintainers in fixing x over 11,000 vulnerabilities over the past eight years as part of the Google Open Source Security Team.
However, the 26 newly identified vulnerabilities are among the first to be detected by OSS-Fuzz with the help of generative AI.
Specifically, the Google researchers used a framework based on a large language model (LLM) trained in-house to generate more fuzz targets.
Fuzz testing, also known as fuzzing, is the most common way developers use to test software for vulnerabilities and bugs before they go into production.
The method involves providing invalid, unexpected or random data as inputs to a computer program or software. The program is then monitored for exceptions such as crashes, failing built-in code assertions, or potential memory leaks.
Fuzz targets are the specific areas of a program or system that are being tested or "fuzzed" by a fuzzer.
On the heels of @Google’s ‘Big Sleep’ AI discovery of a real-world vulnerability, our OSS-Fuzz team identified and reported 26 vulnerabilities to open-source project maintainers by using AI-generated and enhanced fuzz targets. Read more here: https://t.co/6SfPt38ZmE
— Heather Adkins - Ꜻ - Spes consilium non est (@argvee) November 20, 2024
LLM for Automating Fuzzing
With Google’s framework, created in August 2023 and made open source in January 2024, the OSS-Fuzz researchers want to automate the manual process of developing a fuzz target.
This process includes the five following steps:
- Drafting an initial fuzz target
- Fixing any compilation issues that arise
- Running the fuzz target to see how it performs, and fixing any obvious mistakes causing runtime issues
- Running the corrected fuzz target for a longer period, and triaging any crashes to determine the root cause
- Fixing vulnerabilities
The researchers have managed to use their in-house LLM to automate the first two steps and are working on implementing the two following steps.
Google Project Zero AI-Driven Approach for Vulnerability Research
In parallel, researchers at Project Zero, another Google research team, have also built a framework using LLMs for vulnerability discovery.
However, their approach is not built on enhanced fuzzing like the OSS-Fuzz team, but rather centered around the interaction between an AI agent and its set of specialized tools designed to mimic the workflow of a human security researcher and a target codebase.
This project, dubbed Big Sleep, relies on the Naptime framework.
The Project Zero researchers announced in early November 2024 that they found their first vulnerability using this method.
While the researchers conceded that overall, fuzzing will continue to be as – or more – effective as LLM-assisted manual vulnerability analysis, they hope the Big Sleep approach can complement fuzzing techniques by detecting vulnerabilities where fuzzing failed.
Read now: How to Disclose, Report and Patch a Software Vulnerability