AI pentesting agent finds OTP bypass on insurer's live site

An AI pentesting agent found a critical authentication bypass on a major insurer's production website during an authorised security assessment, according to a report published by The Hacker News on 14 October 2026. The flaw sat in a single login request, which let the person signing in choose where their one-time passcode was sent. A one-time passcode, or OTP, is the short code a site sends to a phone or email address to confirm that the person logging in is who they claim to be. If an attacker can pick the destination, the check stops proving anything.

The report says the agent found the issue on its own. The insurer is not named in the source, and no vendor is identified for the agent itself. The source does not say how long the flaw had been present, whether anyone had exploited it, or how many customers used the affected login. It describes the finding as a critical authentication bypass on a live site, not a test environment.

The framing of the report is that the exploitation was straightforward and the validation was not. In the words of the piece, the attack was the easy part, while proving the problem on a production website was the hard part. That distinction matters for anyone running automated offensive tooling against systems that real customers are using at the same moment. Demonstrating an authentication bypass means handling someone's login flow, and doing so without disrupting service or touching accounts that are not part of the agreed scope.

The underlying cause is recorded as a misconfiguration rather than a flaw in a third-party component. In other words, the login endpoint accepted a parameter it should not have trusted. Problems of this shape do not require unusual skill to abuse once they are visible, which is part of why the finding was rated critical.

The status of the issue is listed as reported. The source does not state whether the insurer has fixed the login request, when a fix might land, or whether the company has commented. Nor does it give details of the agent's architecture, the scope agreed with the client, or the controls used to keep the test within that scope.

For security teams, the case is a small data point on a growing question: automated agents are now surfacing serious findings in production, and the governance work around how they prove those findings is lagging the discovery itself.