
Google has discovered an AI-powered hacking system capable of identifying and exploiting website vulnerabilities without human intervention.
In a report published on 16 November, researchers revealed how attackers are using information from security patches to uncover software weaknesses.
It marks the first time Google has identified threat actors using artificial intelligence to fully automate vulnerability discovery and exploitation. Evidence suggests one operator was targeting French websites.
The findings highlight a growing problem for website owners: AI can help hackers exploit security flaws before organisations install the updates intended to fix them.
Google Uncovers AI Hacking System Linked to French Websites
Google Threat Intelligence Group discovered the system, dubbed 'exploit_agent', on an exposed server. Researchers also found data stolen from targeted victims.
The system combines several AI models to identify and exploit weaknesses in web applications. Evidence suggests its operator was particularly interested in French websites associated with media, entertainment and education.
While humans remain involved in directing cyberattacks, Google's discovery demonstrates that certain stages can now run independently.
How Security Patches Could Give AI Hackers an Advantage
Security patches are designed to fix software vulnerabilities. However, changes introduced by an update can also reveal weaknesses in earlier versions.
Google found that attackers are using AI to analyse these changes and develop attacks against vulnerable software.
Developing such attacks has traditionally required technical expertise and time. Automation could leave organisations with considerably less time to protect their systems after developers release security fixes.
AI-Assisted Hackers Breached 11 Organisations in 26 Seconds
A separate investigation by cybersecurity firm GreyNoise documented an AI-assisted campaign targeting PaperCut NG/MF, software used to manage printing.
The operation began on 31 August 2026. Attackers compromised at least 440 software installations across 48 countries, involving 395 identified organisations. The UK accounted for 59 compromised installations.
On 1 September, attackers gained unauthorised access to at least 11 organisations in 26 seconds. In one case, they secured domain administrator privileges at a US high school within seven minutes of gaining initial access.
A human operator directed the AI agents involved. GreyNoise could not establish the attacker's ultimate objective and observed domain administrator access at only 12 victim organisations.
Britain's National Cyber Security Centre Warns of Growing AI Threat
Britain's National Cyber Security Centre (NCSC) has warned that AI will almost certainly increase the frequency and intensity of cyber threats.
Its assessment of threats through 2027 anticipates a widening divide between systems capable of keeping pace with AI-enabled attacks and those that remain more vulnerable.
The agency's 2026 annual review also warns that defenders face mounting pressure to deploy security fixes before attackers can exploit vulnerabilities.
Should You Still Install Security Updates?
The answer is yes; installing security updates remains essential to protecting websites against known software vulnerabilities.
Website owners should prioritise critical patches, maintain an inventory of installed software and replace applications that no longer receive security support.
Automatic updates, where appropriate, can help reduce delays. The danger lies in leaving vulnerable software unpatched, not in installing the updates designed to fix it.
For businesses, schools and other organisations, delaying essential security updates could give AI-assisted attackers the opportunity they need to break in.











