Cybercriminals Using AI-Developed Zero-Day Exploit
New findings reveal that cybercriminals have leveraged a zero-day security exploit, thought to be created with the assistance of artificial intelligence, according to Google. This development marks a notable shift in the landscape of cyber threats.
Google revealed on Monday that its researchers have uncovered what is likely the first documented instance of hackers utilizing a zero-day exploit generated through AI technology. This insight, shared in a report from the Google Threat Intelligence Group, highlights a concerning trend in cybersecurity.
Zero-day exploits are particularly dangerous because they’re unknown to security firms, making them challenging to fix. Such vulnerabilities hold substantial value for cybercriminals and state-sponsored hacking groups, as these flaws can be exploited before developers have a chance to release patches.
This report is the first evidence suggesting that AI has been directly involved in finding and developing such vulnerabilities. This news comes as major AI companies, such as Anthropic and OpenAI, are testing advanced models capable of identifying and exploiting critical software weaknesses, often exceeding the capabilities of most human researchers.
Interestingly, the zero-day exploit in question likely wasn’t developed using Anthropic’s Claude Mythos model, despite its impressive performance in identifying thousands of vulnerabilities across major operating systems and web browsers.
Both the Mythos model and OpenAI’s recent GPT-5.5-Cyber model have garnered significant attention from the Trump administration. Ongoing discussions between government officials and industry groups are focused on potential regulatory frameworks for these advanced AI models.
Prior to the public report, Google informed affected companies about the vulnerability, allowing them to work on and release necessary patches—an approach aimed at responsible disclosure within the cybersecurity field.
John Hultquist, the chief analyst at Google Threat Intelligence Group, underscored the importance of these findings. “For every zero-day that can be traced back to AI, there will probably be many more,” he noted. Hultquist emphasized that this discovery clearly indicates that the race to utilize AI for uncovering network vulnerabilities is already underway. He pointed out, “Attackers are leveraging AI to enhance the speed, scale, and sophistication of their attacks.”
Previously, there have been reports indicating that Anthropic was investigating unauthorized access to its powerful Mythos AI.
This situation raises significant concerns regarding Anthropic’s ability to safeguard its advanced technology from potential misuse. The firm has limited access to Claude Mythos Preview to a select group of tech companies, citing fears that the model could be exploited for large-scale cyberattacks. Reports suggest that an individual gained unauthorized access while acting as a contractor for Anthropic. The company stated there was no evidence of any activity extending beyond a controlled vendor environment, which is typically used by third parties for accessing systems for model development.
