OpenAI's GPT-5.6 Sol identified two critical PHP vulnerabilities, both scored 9.2 on the CVSS scale, capable of unauthenticated remote code execution on servers. Wiz independently confirmed both flaws. The discoveries were made under OpenAI's Patch the Planet initiative, and a coalition of maintainers, researchers, and security providers spent two months preparing fixes before planned public disclosure, because the remediation required a major change to the PHP engine. A separate flaw in Node.js was also found through AI-assisted research.
The embargo on the PHP fix broke more than six hours early, and Amazon observed active exploitation as early as October 17, four days before public disclosure. Government websites and the Internet Archive went offline for emergency patching. The Internet Archive used Claude to create fixes for older PHP versions and local models to run simulated attacks, restoring service quickly. Analysis found that 46.1% of the top 7.2 million PHP websites run end-of-life PHP branches and will not receive official fixes. One publication described the day as "probably the worst day in cybersecurity history. So far.". The premature disclosure may have permanently damaged some long-tail websites.
OpenAI described the episode as an example of what it calls the Defender's Window: "to our knowledge, frontier AI was able to find this vulnerability ahead of attackers, giving defenders time to prepare". OpenAI stated these bugs are "among several significant vulnerabilities we've already found" and expects similar findings to increase sharply in coming months. OpenAI released a Codex skill so website owners can check whether their site is affected.
The discoveries also surfaced a debate about whether AI-assisted defense can sustainably outpace offense. The critical vulnerability was found using Sol and Argon rather than Glasswing, a dedicated defensive program, in the six-plus months since Mythos-level models became available. Defenders did find the bug first, on August 1, but exploitation began October 17, an 11-week gap that drew scrutiny. Nate Soares argued Glasswing's failure to find the bug for six-plus months is strong evidence against the theory that defense can outpace offense on relevant timescales. Vitalik Buterin argued cybersecurity will ultimately be defense-favoring but that the transition period is genuinely hard, and that longer-term defensive advantage requires deeper methodological shifts including formal verification and aggressive trusted-computing-base minimization, which are only beginning. Security experts also warned that attackers now have access to AI cyber capabilities via open-weight Chinese models.