Websites around the world taken offline after AI finds ‘critical’ security flaw

Cyber security chiefs have warned UK organisations their websites may have to be taken offline, after a researcher using AI discovered a “critical security issue” in a programming language.

The Finnish government’s website was one of several to be partly disabled because of the incident on Wednesday, along with the websites of several local authorities in the UK.

Oxford City Council and the City of Edinburgh Council were among the organisations to take their websites offline, while pages for some BBC programmes were unavailable.

The Office for Budget Responsibility said it had followed guidance to take its website offline, “in view of the critical security issues affecting PHP”.

It is understood some of the Government’s online services may be temporarily unavailable while steps are taken to maintain security, but any impact is expected to be minimal.

No large public-facing critical services are thought to have been affected.

The security issue identified by the researcher could allow attackers to remotely control any web servers running the widely used programming language, PHP.

PHP is used by almost 70% of websites, according to cybersecurity firm Cloudflare, and underpins popular platforms such as WordPress and Magento.

According to Cloudflare chief executive Matthew Prince, there was a dramatic surge in attempts to exploit the issue once it was announced.

He wrote on X: “Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it.

“That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.”

Organisations have been urged by the National Cyber Security Centre (NCSC) to update to a new, fixed version of the software.

If software cannot be immediately upgraded, organisations should consider taking their websites offline, the NCSC said.

A Government spokesperson said: “Work is already under way across Government to actively identify and patch any potential exposure.

“We encourage all organisations, private and public, large and small, to follow guidance issued by the NCSC to understand their exposure and take the necessary actions to safeguard their systems.”

The cyber division of the National Security Agency in the US warned weaknesses in the software could provide “a direct path to the server’s file system or database”.

The Wall Street Journal reported that the flaw was discovered in August by Ryan Austin, a researcher at the cybersecurity company Calif, using an advanced AI model developed by OpenAI.

PHP developers had reportedly planned a security update on Thursday, but the release was brought forward after details of its work to address the vulnerabilities leaked online.

OpenAI, which runs ChatGPT, said in the past two months it has been leading a team of researchers working to address the critical flaw in PHP, and supported its developers as it led the fixes.

All that sort of scamming and fraud and so forth gets turbochargedAll that sort of scamming and fraud and so forth gets turbocharged

The company described the incident as “one of the most impactful security issues in recent memory”, adding that while AI can help to uncover issues, it can also aid attackers.

“To our knowledge, frontier AI found these vulnerabilities before attackers did, opening a defender’s window for maintainers and security providers to prepare fixes and protections ahead of disclosure,” it said.

Former head of the NCSC, Ciaran Martin, told Times Radio the vulnerabilities in PHP had the potential to create a significant loss of personal data.

He said: “You’re talking about the potential to take the personal details of cards, including those crucial three digits on the back and so forth, so people will want to be on the alert for scams in future.

“There’s more scope for that. You’re talking potentially for ransomware attackers, not to be able to do ransomware directly as a result of this, but to get a foothold into the organisation.

“You’re talking about easier to impersonate CEOs because you get to know them better. All that sort of scamming and fraud and so forth gets turbocharged.”