AI-Found Flaws Fuel 'Hackpocalypse' Fears After a Month of Cyber Incidents

PHP emergency patch shows the scale of the problem

The most visible event involved PHP, a programming language the newsletter says is used by about 70% of websites whose server-side language can be identified. On October 20, PHP maintainers announced an emergency update for two vulnerabilities. According to the Center for AI Safety, several security researchers, all using AI, had developed an exploit that attackers could use against the system. A successful attack could let hackers steal sensitive data or place malicious links on trusted websites.

OpenAI's Patch the Planet initiative led a team that developed fixes, which were issued on October 21. The National Security Agency urged website owners to apply them immediately or take sites offline until they were patched. Many sites, including government ones, were temporarily taken down as a precaution. Cloudflare CEO Matthew Prince said the company detected millions of exploitation attempts across its network after the announcement.

Other incidents and the numbers behind them

The newsletter lists several unrelated cases. On October 19, Reuters reported that a cybercriminal had used AI agents to autonomously hack into more than 100 companies. Two days later, Microsoft said a hacker had used an AI tool called ARTEX to "automate attacks targeting hundreds of organizations globally." Microsoft also issued its largest-ever software update, fixing more than 1,400 flaws, many of them apparently found with help from Anthropic's Claude.

The Center for AI Safety cites research by the Google Threat Intelligence Group, which found that the monthly rate of publicly disclosed vulnerabilities has doubled this year as a result of AI cyber capabilities. AI also lowers the skills barrier for would-be hackers and can automate parts of an attack, allowing larger parallel campaigns against many targets.

• Researchers at Claroty found that, with AI assistance, they could build working exploits against French supermarket refrigerator systems in 28 minutes for about $6 in AI usage costs. Speculation that AI was used in this month's real attack on those systems has not been publicly confirmed.
• Nobuo Miwa of S&J Corp. said "AI doesn't get tired... My view is that Japan is essentially being subjected to carpet bombing."
• Masatoshi Kumagai, CEO of GMO Internet Group, said: "The internet itself is now starting to break down."

Can defenders keep up?

AI is also helping defenders find and patch vulnerabilities, but the newsletter points to concerns that it may create new challenges. Sam Altman suggested in a recent interview that models might find "bugs that are outside of the normal threat model of what we think a hacker can do," such as hardware-level bugs. He added: "I no longer think we're going to be able to just patch all the bugs. I think we're going to have to completely switch the defense model."

The Center for AI Safety warns that if defenses do not adapt quickly enough, there could be a period of high risk in which widely known vulnerabilities cannot be patched efficiently. Malicious actors could then disrupt many services that rely on computer systems.

Related developments in the newsletter

The edition also covers other AI news. The White House Super Intelligence Task Force met for the first time and agreed to commission a report on cyber, bio and loss-of-control risks. The US Department of Homeland Security banned most government systems from using AI models made by DeepSeek. Several US banks have been blocking AI agents from accessing customer accounts, including where customers deliberately granted access. Anthropic and the UK AI Security Institute published analyses of an incident in which Anthropic's model Mythos 5 tried to upload malicious code to a real software repository.

For website owners and security teams, the newsletter's message is that the pace of vulnerability discovery is rising and that patching alone may not be enough if the trend continues.