Google Says Its Ai-based Bug Hunter Found 20 Security Vulnerabilities

Trending 1 month ago

Google’s AI-powered bug huntsman has conscionable reported its first batch of information vulnerabilities. 

Heather Adkins, Google’s vice president of security, announced Monday that its LLM-based vulnerability interrogator Big Sleep recovered and reported 20 flaws successful various celebrated unfastened root software.

Adkins said that Big Sleep, which is developed by nan company’s AI section DeepMind arsenic good arsenic its elite squad of hackers Project Zero, reported its first-ever vulnerabilities, mostly successful unfastened root package specified arsenic audio and video room FFmpeg and image editing suite ImageMagick. 

Given that nan vulnerabilities are not fixed yet, we don’t person specifications of their effect aliases severity, arsenic Google does not yet want to supply details, which is simply a modular argumentation erstwhile waiting for bugs to beryllium fixed. But nan elemental truth that Big Sleep recovered these vulnerabilities is significant, arsenic it shows these devices are starting to get existent results, moreover if location was a quality progressive successful this case. 

“To guarantee precocious value and actionable reports, we person a quality master successful nan loop earlier reporting, but each vulnerability was recovered and reproduced by nan AI supplier without quality intervention,” Google’s spokesperson Kimberly Samra told TechCrunch. 

Royal Hansen, Google’s vice president of engineering, wrote connected X that nan findings show “a caller frontier successful automated vulnerability discovery.” 

LLM-powered devices that tin look for and find vulnerabilities are already a reality. Other than Big Sleep, there’s RunSybil, and XBOW, among others. 

Techcrunch event

San Francisco | October 27-29, 2025

XBOW has garnered headlines aft it reached nan top of 1 of nan U.S. leaderboards astatine bug bounty level HackerOne. It’s important to statement that successful astir cases, these reports person a quality astatine immoderate constituent of nan process to verify that nan AI-powered bug huntsman recovered a morganatic vulnerability, arsenic is nan lawsuit pinch Big Sleep.

Vlad Ionescu, co-founder and main exertion serviceman astatine RunSybil, a startup that develops AI-powered bug hunters, told TechCrunch that Big Sleep is simply a “legit” project, fixed that it has “good design, group down it cognize what they’re doing, Project Zero has nan bug uncovering acquisition and DeepMind has nan firepower and tokens to propulsion astatine it.”

There is evidently a batch of committedness pinch these tools, but besides important downsides. Several group who support different package projects person complained of bug reports that are really hallucinations, pinch immoderate calling them nan bug bounty balanced of AI slop. 

“That’s nan problem group are moving into, is we’re getting a batch of worldly that looks for illustration gold, but it’s really conscionable crap,” Ionescu antecedently told TechCrunch.

Lorenzo Franceschi-Bicchierai is simply a Senior Writer astatine TechCrunch, wherever he covers hacking, cybersecurity, surveillance, and privacy. You tin interaction Lorenzo securely connected Signal astatine +1 917 257 1382, connected Keybase/Telegram @lorenzofb, aliases via email astatine lorenzo@techcrunch.com.

More