The PHP Meltdown: How AI Makes Software Bugs More Exploitable

Security teams and website administrators rushed to patch two critical PHP vulnerabilities this week. Within hours of their public disclosure, Cloudflare recorded millions of attempts to exploit them. The flaws, known collectively as PHP2Shell, affect a widely used web scripting language, making them an appealing target for attackers.

So far, the feared wave of successful intrusions has not materialized. The PHP Foundation says it has seen no confirmed successful exploitation, and some exploits circulating online appear not to work. Still, the episode has highlighted how AI can turn old vulnerabilities into urgent threats: researchers used advanced AI tools to develop practical exploits for weaknesses that had been known for years.

Researchers at Calif first identified the flaws and built working exploits in August, using frontier AI tools as part of OpenAI’s Patch the Planet program. Their work also depended on human expertise and a detailed understanding of PHP. Researchers at OpenAI and security company Wiz later found the vulnerabilities independently with AI assistance.

Some of the underlying issues had been identified as early as 2024, but were considered too difficult to exploit at the time. Wiz researcher Sagi Tzadik says the PHP team had known about the vulnerability for years but had not addressed it because it did not appear usable by attackers—at least, not by humans working alone.

AI is accelerating the search for vulnerabilities and the development of ways to exploit them, adding pressure on developers and overstretched security teams. Microsoft’s October security updates addressed a record 1,453 vulnerabilities, while Bank of America reportedly hired 1,000 consultants to fix problems identified by Anthropic’s Mythos.

However, researchers caution that AI is amplifying familiar security dynamics rather than introducing an entirely new kind of threat. Jake Williams, a former NSA hacker, expects AI to uncover more long-standing flaws but says PHP is unusually exposed: unlike many tools, it is designed to be accessible over the internet.

Sophisticated AI-assisted hacking still generally involves skilled people. Ryan Austin, the Calif researcher who first found the PHP flaws, notes that no model can uncover every bug simply by receiving a prompt. Security researcher Niels Provos is similarly unconvinced that AI will suddenly expose an overwhelming number of old or previously impractical vulnerabilities. Finding PHP2Shell, he says, required a highly skilled team—even after researchers had been using AI for more than a year.

Total
0
Shares
Leave a Reply

Your email address will not be published. Required fields are marked *

Previous Article

The PHP Meltdown: How AI Is Making Software Bugs More Exploitable

Next Article

What the U.S. Movement Against Data Centers Can Learn From Abroad

Related Posts