Google mentioned it found a zero-day vulnerability in the SQLite open-source database engine utilizing its giant language mannequin (LLM) assisted framework known as Big Sleep (previously Mission Naptime).
The tech large described the event because the “first real-world vulnerability” uncovered utilizing the unreal intelligence (AI) agent.
“We consider that is the primary public instance of an AI agent discovering a beforehand unknown exploitable memory-safety situation in extensively used real-world software program,” the Big Sleep crew said in a weblog publish shared with The Hacker Information.
The vulnerability in query is a stack buffer underflow in SQLite, which happens when a bit of software program references a reminiscence location previous to the start of the reminiscence buffer, thereby ensuing in a crash or arbitrary code execution.
“This usually happens when a pointer or its index is decremented to a place earlier than the buffer, when pointer arithmetic outcomes in a place earlier than the start of the legitimate reminiscence location, or when a adverse index is used,” based on a Frequent Weak point Enumeration (CWE) description of the bug class.
Following accountable disclosure, the shortcoming has been addressed as of early October 2024. It is value noting that the flaw was found in a improvement department of the library, which means it was flagged earlier than it made it into an official launch.
Mission Naptime was first detailed by Google in June 2024 as a technical framework to enhance automated vulnerability discovery approaches. It has since advanced into Big Sleep, as a part of a broader collaboration between Google Mission Zero and Google DeepMind.
With Big Sleep, the concept is to leverage an AI agent to simulate human conduct when figuring out and demonstrating safety vulnerabilities by profiting from an LLM’s code comprehension and reasoning skills.
This entails utilizing a collection of specialised instruments that permit the agent to navigate by way of the goal codebase, run Python scripts in a sandboxed setting to generate inputs for fuzzing, and debug this system and observe outcomes.
“We predict that this work has large defensive potential. Discovering vulnerabilities in software program earlier than it is even launched, implies that there is no scope for attackers to compete: the vulnerabilities are fastened earlier than attackers actually have a probability to make use of them,” Google mentioned.
The corporate, nonetheless, additionally emphasised that these are nonetheless experimental outcomes, including “the place of the Big Sleep crew is that at current, it is probably {that a} target-specific fuzzer could be at the least as efficient (at discovering vulnerabilities).”