From one PHP setup to many
An exploit built for a specific PHP installation used to require substantial adaptation before it could work across other versions and configurations. Wiz says large language models can automate that work, adjusting exploits to different PHP versions, configurations and memory layouts.
That changes the practical reach of a vulnerability. A flaw that was difficult to exploit on one system can become easier to adapt across many systems.
The fix took two months and involved five stages. PHP developers, Wiz, Cloudflare and US government bodies, including CISA and the White House, took part. OpenAI presents the process as a framework that could be repeated for future AI-assisted cybersecurity discoveries.
The question beyond the discovery
OpenAI expects the number of discoveries on the scale of PHP2Shell to rise sharply in the coming months. It also warns that the same AI capabilities that help researchers find vulnerabilities could help attackers.
I think the most important claim here is not that AI found a flaw, but that it can make exploit adaptation less labor-intensive. That could widen the pool of systems exposed to a known vulnerability. The announcement says the fix was checked, but offers little detail on how the detection tools performed or how widely the repaired versions have been adopted.
If AI makes vulnerability research faster, the advantage will depend not only on who finds a flaw first, but on whether defenders can distribute and verify a fix before the same capabilities reach attackers.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X