AI-powered hacking wave overwhelms responders as web flaw hits 70% of sites

Summary

• Hackers are deploying AI agents and automated tools at scale, flooding the internet in recent weeks.
• Newly disclosed web bugs could allow takeover of as many as 70% of the world's websites.
• Incident-response firms expect a massive number of breach disclosures over the coming weeks and months.

The latest

A wave of AI-powered intrusions has swamped cybersecurity responders this month, with hackers using autonomous agents and automated tools that make attacks faster and harder to trace. CrowdStrike senior vice president Adam Meyers said both advanced and unsophisticated attackers are now running what he described as agentic adversaries, according to The Wall Street Journal.

Details

• The flaw: Organizations scrambled Wednesday to patch newly uncovered web bugs that could allow takeover of as many as 70% of the world's websites, according to The Wall Street Journal. Amazon.com said hackers are already actively trying to break into sites by leveraging the flaws.
• The government warning: The National Security Agency and the Federal Bureau of Investigation told companies unable to patch immediately to pull the plug on their websites. Government-run sites in Canada, Belgium, France and the Netherlands did exactly that, and the message board 4chan suspended comments ahead of the patches.
• The tool: One instrument in circulation is Artex, a Chinese open-source hacking tool that uses artificial intelligence to automate breaking into a computer network. Microsoft said this week that threat actors across multiple regions were using Artex to accelerate attacks on hundreds of companies.
• The casualties: A cyberattack earlier this month used Artex to steal data on about 68,000 people from South Korean banks, according to The Wall Street Journal. Incident responders say AI tooling is now in the hands of nation-state hackers from China and Russia alongside criminals seeking data and ransomware payouts.
• The individual case: Startup founder Colin Parsons was billed $449 after hackers used credentials he had stored with AI-coding site OpenCode to buy Anthropic and OpenAI services, starting at 11 p.m. on a recent Friday. He woke to roughly 50 charge notifications.
• The platform: OpenCode has been locked in what co-founder Dax Raas called a non-stop AI-fueled cyberattack since Oct. 16. Raas wrote on X that the company essentially needs agents running around the clock because that is what the attackers have. OpenCode representatives did not reply to messages seeking comment.
• The spillover: Parsons found OpenCode's Discord support channel flooded with nonsensical messages arriving every two seconds, burying his fraud report. Other users replied on X that they had been hit too, one citing $1,700 in charges and another saying his OpenAI account was suspended after his keys leaked.
• The ransomware shift: Sophos researchers gained access to systems used by a Russian-speaking ransomware group calling itself Aurora and watched its AI dependence grow through the year, from building tooling and infrastructure to probing hacked corporate networks for routes into other accounts, said threat intelligence manager Colin Cowie.
• The economics: Aurora shifted from proprietary models such as Anthropic's to open-source ones including Moonshot AI's Kimi models, Cowie said, and accessed them through a $200-per-month license from AI-coding service Cursor. He said the group operates faster than the average human.
• The measurement: LMG Security founder Sherri Davidoff said hackers now use AI for reconnaissance, for analyzing vast troves of stolen data and for rapidly building exploit code. In quarterly polling she runs with NetDiligence, virtually no peers reported AI use a year ago; last quarter it had become routine.

Between the lines

The two threads running through October compound each other: AI is both surfacing security bugs at speed and arming the attackers racing to exploit them. The Aurora case suggests the barrier is cost, not skill — a $200 monthly coding license and open-source models replaced proprietary access, consistent with Meyers's observation that unsophisticated attackers now use the same agentic methods as advanced ones.

What's next

Watch the volume of breach disclosures Volexity president Steven Adair expects over the coming weeks, patch-completion rates on the web flaw, whether more government sites go dark, and whether OpenCode responds publicly to affected users.

Source: The Wall Street Journal