AI Agent Exploits Four Flaws to Hijack London Recruiting Platform in Under an Hour

A single AI agent outsmarted a recruiting platform's defenses—no passwords, no hacking skills, just chained exploits. Could this redefine cybersecurity threats?

, and Administrator

2026 May 10 . 2:02 PM

1 min read

The image shows a cartoon of a man in a police uniform holding a sign that reads "I suspect our AI... — The image shows a cartoon of a man in a police uniform holding a sign that reads "I suspect our AI is plotting something against us" while two robots stand in front of him, one of them holding a paper with text on it. In the background, there is a wall with a screen and buttons.

AI Agent Exploits Four Flaws to Hijack London Recruiting Platform in Under an Hour

An AI security firm claims its autonomous agent uncovered serious flaws in a London-based recruiting platform. Codewall reported that the agent exploited four vulnerabilities on Jack & Jill, gaining full admin rights within an hour. The findings have not yet been independently confirmed. The agent began by creating an account using a third-party corporate email domain. It authenticated via Clerk’s test mode, which relied on a static one-time code. This allowed automatic linking to an existing company profile, granting full admin privileges.

Next, the agent found a missing role verification step during corporate onboarding. It then discovered an endpoint that assigned users to companies based only on email domains—without confirming ownership. A URL fetcher also exposed internal API documentation, accessible without login.

The agent went further by testing the platform’s voice infrastructure. Using text-to-speech, it generated synthetic voice clips and interacted directly with Jack & Jill’s AI assistant, 'Jack'. According to Codewall, these steps formed a chain of exploits leading to complete control over corporate accounts. Codewall’s report suggests the agent achieved full admin access in under an hour. The vulnerabilities involved authentication flaws, missing ownership checks, and exposed internal tools. Independent verification of the findings remains pending.