Google’s cybersecurity team caught a hacking group using artificial intelligence to hunt for a weakness in a company’s digital infrastructure, then stopped the operation before it could do damage. The group, which Google tracks internally as Mythos, had been feeding an AI model information about the target’s systems in an effort to automate vulnerability discovery and scale an exploitation campaign.
Google’s Threat Intelligence Group, known as GTIG, identified the activity, alerted the targeted company, and coordinated with law enforcement to shut the campaign down. John Hultquist, who leads threat analysis at GTIG, described the case as a milestone in the evolution of cyber threats.
How Mythos used AI to accelerate an attack
Traditional hacking campaigns require skilled analysts to comb through source code, documentation, and network configurations looking for exploitable flaws. That process can take weeks or months. AI compresses it. By feeding an AI model data about a target’s systems, attackers can parse vast amounts of technical information at machine speed, flagging potential vulnerabilities and even suggesting attack paths that a human operator can then refine and execute.
That is what Mythos appears to have done. The group used AI specifically during the vulnerability-scanning phase, the stage where attackers identify which doors might be unlocked before they try to walk through them. Google caught the activity before it resulted in a breach, which suggests the defensive side of this arms race is keeping pace, at least for now.
The case is not the first time security researchers have documented threat actors experimenting with AI. In February 2024, Microsoft and OpenAI disclosed that state-sponsored hacking groups linked to Russia, China, Iran, and North Korea had used large language models for tasks like scripting, reconnaissance research, and refining social engineering lures. But those cases involved AI as a productivity tool for hackers. The Mythos operation goes further: it represents AI being used to actively probe for and identify exploitable weaknesses in a specific target’s defenses.
What Google has not disclosed
Key details remain under wraps. Google has not said which AI model Mythos used, whether it was a commercial product, an open-source system, or something the group built or modified on its own. The identity of the targeted company and the nature of the vulnerability being probed are also undisclosed.
The scale and origin of Mythos itself are unclear. No public reporting has placed the group geographically or linked it to a nation-state sponsor. Criminal hacking outfits and government-backed teams frequently share tactics and tooling, and without attribution details, it is difficult to assess where Mythos fits in the broader threat landscape.
Law enforcement involvement has been confirmed in general terms, but no agency has been named and no charges have surfaced. That leaves an open question: did the disruption stop at a technical block and a warning to the target, or did it lead to arrests, infrastructure seizures, or other enforcement actions? The distinction matters for judging how thoroughly the threat was neutralized.
The defensive side of the AI arms race
Google has been building toward this kind of interception for some time. The company’s Threat Intelligence team has published research on how AI systems themselves can be weaponized through prompt injection, a technique where malicious inputs trick an AI model into executing harmful commands. A related academic paper available through arXiv examines methods for estimating the risk of such attacks. Google has not drawn a direct line between that research and the Mythos case, but the work shows the company has been studying both the offensive and defensive dimensions of AI security in parallel.
Google’s own Threat Intelligence Group has also tracked the broader trend of AI adoption by threat actors across its reporting, noting a steady increase in the sophistication of AI-assisted techniques used by both criminal and state-sponsored groups.
Other major technology companies have taken similar steps. Microsoft has integrated AI-driven threat detection into its security products, and OpenAI has published transparency reports on attempts to misuse its models. But the Mythos disruption stands out because it involves AI being used not just for support tasks like writing phishing emails but for the core technical work of finding and preparing to exploit a software flaw.
What this means for organizations and policy
For security teams, the practical implication is straightforward but uncomfortable. AI gives attackers the ability to cycle through potential vulnerabilities faster, pivot between targets more fluidly, and adapt techniques with less effort. The fundamentals of defense still apply: patch management, secure coding, network segmentation, and incident response are not going anywhere. What changes is the speed at which those defenses get tested.
That acceleration may force defenders to adopt their own AI-driven tools for anomaly detection, threat hunting, and automated response just to maintain parity. Google’s success in catching Mythos before a breach occurred is a proof point that AI-powered defense can work, but it is a single case, and the company has not shared enough technical detail for others to replicate its detection methods.
The lack of specifics also creates a problem for the targeted company’s peers. Without knowing which vulnerability was being probed, other organizations cannot audit their own systems for the same weakness or determine whether similar campaigns might already be underway against them. They are left with a general warning about attacker capability rather than a specific indicator they can act on.
For regulators, the Mythos case is likely to accelerate conversations about AI-specific security requirements. Logging of AI model access, red-teaming mandates, and stricter controls on how models can be queried are already moving from best-practice recommendations toward formal expectations in critical sectors. But crafting effective rules requires understanding how attackers actually use AI, and the information gaps in this case make that harder.
Why the Mythos case matters despite its unanswered questions
The Mythos disruption confirms what cybersecurity professionals have been warning about for years: AI is no longer just a subject of conference talks and research papers. It is an operational tool in the hands of attackers. The fact that Google caught this particular campaign is encouraging. The fact that so many details remain hidden is not.
What the case does not prove is equally important. There is no evidence that AI autonomously discovered a zero-day vulnerability without human guidance. Human operators were still behind the keyboard, shaping prompts, interpreting outputs, and deciding how to turn AI-generated insights into actionable exploits. The breakthrough is in speed and scale, not in replacing human expertise.
Until more cases like this are documented with comparable or greater detail, security professionals will have to navigate the uncertainty by strengthening core defenses, experimenting with their own AI capabilities, and operating under the assumption that adversaries are already doing the same.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
