AI's hacking capabilities are rapidly advancing, reaching a critical point. Cybersecurity startup RunSybil discovered its AI tool, Sybil, identified a previously unknown vulnerability in a customer's system in November. Sybil, which uses AI models to scan for exploitable weaknesses, detected a flaw in the customer's federated GraphQL deployment. This exposed confidential information.

The cofounders of RunSybil, Vlad Ionescu and Ariel Herbert-Voss, were surprised by the AI's ability to identify the complex issue. They confirmed no public record of the vulnerability existed. Sybil uses a combination of AI models and proprietary techniques to find security flaws like unpatched servers and misconfigured databases.

RunSybil has since found the same GraphQL problem in other systems before public disclosure. Experts believe this demonstrates a significant leap in AI's reasoning and hacking abilities. The discovery highlights the urgent need for enhanced AI-driven cybersecurity defenses.

GraphQL is a language that specifies how data is accessed over the web through APIs. RunSybil's Sybil aims to proactively identify such vulnerabilities before malicious actors can exploit them. The company plans to continue refining Sybil's capabilities to stay ahead of evolving AI hacking techniques.