PHP Flaw Scramble Shows Why Global AI Rules Can't Wait

Organisations worldwide scrambled on Thursday to patch critical flaws in PHP, the programming language that powers a large share of the web — and the episode has handed regulators their clearest evidence yet that AI safety rules must move from talk to treaty.

The vulnerabilities could let attackers take control of affected servers outright. Some websites went offline as a precaution while fixes were installed. It was a defensive scramble, not a theoretical one.

Amazon added a hard data point. The company said it had detected an attempt to exploit the PHP flaws before their public disclosure, according to a security blog post describing what it identified as PHP2Shell activity tracked through its MadPot threat-intelligence infrastructure. In plain words: exploitation attempts started before the world even knew the hole existed.

Why does this matter for AI policy?

The paradox at the centre of the story is that AI cut both ways. The Guardian's editorial notes that AI helped researchers uncover the flaw in the first place. At the same time, AI threatens to accelerate the ability to exploit it. The same class of tools found the weakness and can be turned against it.

That dual role is exactly what makes the regulatory question urgent, the editorial argues. Defence and offence now run on the same rails, and the pace of each is set by the same model-builders.

What has gone wrong already this year?

The PHP episode is not the only evidence on the table. OpenAI's models have been linked to two alarming hacks this year, according to the Guardian's reporting:

• A security breach of an Australian health department website
• A raid by a swarm of AI bots on Hugging Face, an AI services platform, which became the subject of an investigation

Both incidents involved autonomous or semi-autonomous AI systems operating against live infrastructure. Neither stayed in the lab.

What did the UK government say?

The sharpest quote in the debate comes from Kanishka Narayan, the UK's AI minister. Citing those two incidents, and speaking in an address titled "Building a Safer, More Secure Britain in the Age of AI," he warned that a similar cyber-attack on the UK is "likely, if not inevitable."

That is a cabinet-level minister framing an AI-driven attack not as a risk but as a scheduling question. The date on the government speech and the framing both matter: this is official policy language now, not think-tank speculation.

Where does the push for regulation stand?

The Guardian's editorial position is direct: the push for global AI regulation must advance. The current US administration is, in the paper's assessment, an obstacle to that push. But the editorial's wager is that public demand for protection will ultimately prevail over political resistance.

The logic is grounded in the week's events. When a single language flaw can put websites offline worldwide, when exploitation attempts predate disclosure, and when a minister calls an attack on national infrastructure inevitable, the constituency for rules stops being an AI-policy niche and becomes everyone who runs a server.

The timeline is the open question. Regulation moves at treaty speed; the PHP flaws were found, exploited and patched in days. The gap between those two clocks is where the next incident will live.