The News Central

AI Tools Are Finding Software Flaws Faster Than Volunteer Maintainers Can Fix Them, Developers Warn After Emergency PHP Patches

Browser makers and open-source foundations report record numbers of newly found bugs, and the people maintaining widely used code say they cannot keep up.

Earlier this month governments and companies around the world urged website administrators to install emergency updates for PHP, the programming language behind a large share of the web. Independent security researchers had used AI tools to find critical flaws that had sat in the code for more than a decade and that humans would have found too hard to exploit. Some government and corporate sites briefly went offline while they were patched. The episode, known as PHP2Shell, may be the first of many, say the people who maintain open-source software.

PHP is said to run roughly 70 percent of websites. It is coordinated by a non-profit foundation set up in 2021 with a small budget.

Record bug counts

The numbers are rising across the industry. Google's latest Chrome update fixed more than 1,000 security bugs, more than the previous two dozen updates combined. Mozilla found over 400 vulnerabilities in April, against a monthly record of 76 before that. Jarek Potiuk, who sits on the volunteer security team of the Apache Software Foundation, says the foundation now has about ten times its usual number of issues because it is using AI to look for them.

Finding flaws is only part of the work. AI scans cost money in processing charges, and people must review the results to weed out false alarms. Potiuk says every project and foundation is now spending heavily on this. Amanda Brock, chief executive of an open-source advocacy group, says it is not possible with present resources to fix all known vulnerabilities, and that the world, including commercial software vendors, depends on this shared code.

Industry response

Anthropic launched Project Glasswing earlier this year with $100 million in credits so that its Claude Mythos model could search for vulnerabilities. In its first month the programme reportedly found more than 23,000 flaws across over 1,000 open-source projects. A separate scheme, Alpha-Omega, backed by Anthropic, Amazon, Google, Microsoft and OpenAI, spends about $7 million a year helping developers find critical bugs, and the Apache foundation uses it to scan its 290 packages.

Potiuk argues that large companies, which depend on free software worth trillions in market value, should pay their own staff to spend working hours maintaining the projects they use. He calls the present imbalance highly disproportionate. Peter Slattery, a researcher at MIT who tracks real-world AI-related cyber incidents, says detection and patching are improving but attacks are spreading too, and that it is a matter of time before some of the newly found flaws are exploited. Potiuk points to the 2021 Log4Shell flaw, found without AI, which let attackers run code on millions of devices.

The open question is who pays for the fixing.