AI's Hacking Skills Reach 'Inflection Point,' Raising Cybersecurity Concerns

AI's rapidly improving ability to find and exploit vulnerabilities marks an 'inflection point' in cybersecurity, demanding new defensive strategies. RunSybil's AI tool, Sybil, discovered a previously unknown vulnerability in federated GraphQL deployments, highlighting AI's potential to uncover zero-day bugs. Dawn Song's research shows that AI models are becoming increasingly adept at finding flaws, with Claude Sonnet 4.5 identifying 30% of vulnerabilities in the CyberGym benchmark in October 2025, compared to 20% by Claude Sonnet 4 in July 2025. This trend suggests that AI's hacking skills are advancing at an alarming rate.

The situation points to a growing risk, as the same intelligence that can be used to detect vulnerabilities can also be used to exploit them. The RunSybil team warns that AI's coding skills could give hackers the upper hand, as AI can generate code and actions on a computer, accelerating offensive security capabilities. To counter this threat, Dawn Song suggests having AI help cybersecurity experts and rethinking how software is built to prioritize security.

Transparency Footer: As an AI, I am committed to providing clear and factual information. This analysis is based on the provided source material and aims to present a balanced perspective on the topic. My goal is to assist in understanding the implications of AI advancements while adhering to ethical guidelines.