AI Cyber-Hound: Google’s Big Sleep AI Finds 5 Critical Flaws in Apple’s WebKit
Google has developed an artificial intelligence system called Big Sleep, which is already proving valuable in the field of cybersecurity. This week, Apple formally thanked the company for identifying five vulnerabilities in WebKit, the browser engine used by Safari. These flaws could have caused browser crashes or memory corruption when processing specially crafted web pages.
Big Sleep is a large language model–based system created through collaboration between DeepMind and Google Project Zero. Initially codenamed Project Naptime, it was conceived as a tool for automated vulnerability discovery. The system employs machine learning algorithms to analyze code and detect potentially dangerous constructs that could lead to crashes or data leaks.
Apple confirmed that the vulnerabilities were patched in updates released on November 3, covering all of its operating systems and the Safari browser. Fixes were deployed to iOS 26.1, iPadOS 26.1, macOS Tahoe 26.1, tvOS 26.1, watchOS 26.1, visionOS 26.1, and Safari 26.1. The updates are available for iPhone 11 and later, iPad Pro 12.9-inch (3rd generation and newer), iPad Air (3rd generation and above), iPad mini (5th generation and above), Apple Watch Series 6 and newer, Apple TV 4K (2nd generation and above), and the Vision Pro headset.
In total, Google Big Sleep helped uncover the following five vulnerabilities:
- CVE-2025-43429 — A buffer overflow issue that could cause process crashes when handling malicious content. Fixed by improving memory boundary checks.
- CVE-2025-43430 — An unspecified error leading to browser crashes when opening malicious web pages. Addressed through enhanced state management.
- CVE-2025-43431 and CVE-2025-43433 — Two independent memory corruption flaws triggered by crafted data inputs. Apple mitigated these through improved memory handling.
- CVE-2025-43434 — A use-after-free vulnerability that could cause Safari to terminate unexpectedly. Resolved via strengthened state control mechanisms.
Google emphasized that Big Sleep has already demonstrated its effectiveness: earlier in 2025, the system detected a severe vulnerability in the SQLite library (CVE-2025-6965) rated 7.2 on the CVSS scale — a flaw considered to pose a credible exploitation risk.
Although Apple reported no evidence of these WebKit vulnerabilities being actively exploited in attacks, the company advised all users to install the latest updates promptly to minimize potential risk. The patches are now available through standard software update channels.
Big Sleep is regarded as one of the first fully autonomous systems capable of independently analyzing source code and identifying vulnerabilities by correlating them with known attack patterns. The project exemplifies how artificial intelligence is evolving into a vital instrument — not only for software development but also for safeguarding the digital ecosystems of the world’s leading technology companies.
Support Our Threat Intelligence
If you find our technology report and cybersecurity news helpful, consider supporting our work.