Security researchers have long warned that artificial intelligence would eventually change how cyberattacks are carried out. A new report from Google suggests that point has arrived.
Google’s Threat Intelligence Group says that over the past quarter it identified, for the first time, attackers using large-language-model frameworks to automate vulnerability discovery and exploitation with no human in the loop.
What Google found
The most striking example came from an unexpected place. Google’s analysts found an exposed staging server belonging to attackers that hosted a multi-agent framework, dubbed exploit_agent, which connects to several AI models. The server also held stolen victim data, reconnaissance information and other attack tools.
The system studies software updates for popular web applications, looks for signs of security flaws and then automatically tries to exploit them. Google linked it to attacks on French media, entertainment and education websites.
A shift in scale
According to the report, this is no longer experimental. Google says AI began as a trial tool in attacks but has now been adopted at scale by both financially motivated and state-backed groups. Attackers are increasingly relying on open-source penetration-testing suites. These include ARTEX, CAIRN, CyberStrikeAI and HexStrike AI, which are used against internet-facing systems with easy-to-exploit flaws.
The US government has noticed too. In an advisory last week, the FBI warned that multiple threat actors have been using the ARTEX framework since August against organisations in finance, government, technology, defence and healthcare.
The forecast
Google’s outlook is blunt. Its researchers estimate that within three months attackers will be able to launch unsophisticated attacks that need little or no human interaction, and that sophisticated, fully autonomous attacks are likely to emerge within six months.
Those timelines are projections, not certainties. But the direction is clear, and security teams have little reason to wait.
Less time to patch
The most immediate effect is on speed. Google says the use of such frameworks has sharply reduced the time defenders have to fix vulnerabilities. It also creates noise. After PHP announced two flaws, hundreds of purported proof-of-concept exploits, mostly AI-generated, appeared quickly, and defenders had to check the code and build protections.
For companies, the lesson is that a gap of days or weeks between a patch being published and installed has become riskier. Attackers who can scan for newly disclosed flaws at machine speed will find unpatched systems faster than people can.
Other trends
The report points to wider misuse of AI tools. Google described attacks by a cluster tracked as UNC6836, with suspected links to “Midnight Neptune,” that targeted cryptocurrency firms in North America and Europe using a modified version of the Claude Code assistant, altering its settings so a backdoor was downloaded on launch. The report also says state-backed groups, commercial vendors and political networks now routinely use AI in influence operations, and that “LLMjacking,” or stealing access to AI services, has become widespread and costly for victims.
What organisations can do
Specialists generally recommend a few basics, and they matter more as attacks speed up. Keep an accurate list of internet-facing systems. Apply security updates quickly, prioritising flaws that are being actively exploited. Use multi-factor authentication. Limit what software and AI tools can access. Monitor for unusual activity, and make sure incident-response plans are practised, not just written.
For individuals, the advice is familiar: update devices promptly, use strong unique passwords with a password manager, and treat unexpected messages with suspicion.
The larger point is that defenders will also use AI to find and fix flaws, and some security firms already do. The race is now between automated attack and automated defence, and Google’s warning is a sign that the attackers have started to move.

