ETHERDORF FOUNDATION: AI FOUND LOOPHOLES THAT COULD LEAD TO THE DE-LINKING OF THE CERTIFIER

According to the ETA Foundation, the AI agent recently deployed on its software identified a loophole that could lead to the de-linking of the certificationer. Although loopholes had been identified, a careful human judgement was needed to distinguish between real loopholes and false positives. The loophole is in Gossipsub and allows a remote system to trigger a collapse, leading to an uncalculated node software and then to shut down the certifier until the operator restarts. The loophole was quickly repaired and disclosed as `CVE-2026-34219'. It also noted that AI agents had done well in identifying loopholes, but were weak in addressing gaps that cut across multiple steps。