Researchers used an AI model to discover two longstanding bugs in PHP, code that powers an estimated 70% of websites. One vulnerability, PHP2Shell, could allow attackers to take control of servers without credentials. The bugs remained hidden because PHP runs in so many systems that teams often lose track of where it lives. The exploit details leaked before patches were available, widening the window of risk.
In a separate move, Anthropic has given six outside evaluators employee-level access to test its AI models for safety and capability issues. The company is essentially running a pilot audit program with external parties. The broader question: should frontier AI labs go further and let competitors and regulators directly inspect their systems?
This is an AI-written summary. The original reporting belongs to @IBMTechnology. How we write these · Spot a mistake?