FOLLOW US

The next generation of artificial intelligence-driven cyberattacks could far surpass anything previously encountered, according to warnings from industry leaders and researchers.

That concern was underscored last week when AI company Anthropic outlined risks tied to its upcoming model, Mythos, in a leaked blog post. The draft suggested the system—and others like it—could identify and exploit software vulnerabilities at unprecedented speed.

Anthropic is not alone in sounding the alarm. OpenAI cautioned in December that its forthcoming models could present a “high” cybersecurity risk. Analysts have already warned that artificial intelligence can intensify existing threats while accelerating the creation of new forms of cyber exploits.

The emergence of AI agents—systems capable of autonomously executing tasks—could further amplify those risks. Experts say a single AI agent might scan for and exploit weaknesses faster and more persistently than large groups of human hackers.

“The agentic attackers are coming,” said Shlomo Kramer, founder and CEO of cybersecurity and networking company Cato Networks. “This is a watershed event in the history of cybersecurity.”

Leak Reveals Details of ‘Mythos’

Information about Mythos surfaced in an unpublished blog post first reported by Fortune. Anthropic did not respond to CNN’s request for comment but told Fortune the disclosure stemmed from human error within its content management system.

“Although Mythos is currently far ahead of any other AI model in cyber capabilities, it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders,” Anthropic said in the draft.

The company indicated it is allowing select organizations to test the system in advance, aiming to strengthen defenses “against the impending wave of AI-driven exploits.”

According to Axios, Anthropic has also privately briefed government officials on the possibility of large-scale cyberattacks enabled by Mythos.

Still, some experts argue that the broader trend—not a single model—is the greater concern.

“Behind Mythos is the next OpenAI model, and the next Google Gemini, and a few months behind them are the open-source Chinese models,” Kramer told CNN.

AI is already enabling near-instant exploitation of newly discovered vulnerabilities, said Evan Peña, chief offensive security officer at cybersecurity firm Armadin.

Even so, limitations remain. Peña noted that while advanced AI systems excel at identifying weaknesses and generating exploit code, they lack the contextual awareness human attackers use to determine which data is most valuable.

Human oversight will remain essential in AI-driven cyber operations, said Joe Lin, co-founder and CEO of Twenty, a company providing offensive cyber tools to the U.S. government.

“We must ensure we are building weapons systems where humans remain firmly in control of decisions and outcomes, because while the machine handles the execution, the human must always own the consequences,” he said.

Rise in AI-Assisted Cyberattacks

Recent incidents illustrate how AI tools are lowering the barrier to entry for cybercrime. In January, a Russian-speaking hacker used multiple AI systems to compromise more than 600 devices running widely used firewall software across 55 countries, according to Amazon Web Services’ security research team.

The attacker relied on generative AI to “implement and scale well-known attack techniques throughout every phase of their operations, despite their limited technical capabilities,” AWS reported.

The operation incorporated Anthropic’s Claude model and the Chinese-developed DeepSeek system, said Eyal Sela, director of threat intelligence at Gambit Security. Chat logs reviewed by CNN showed the hacker asking Claude, in Russian, to build a web-based control panel to manage hundreds of targets.

According to Sela, AI effectively grants hackers of varying skill levels “superpowers” by reducing the technical expertise required to breach systems.

In another case, Bloomberg reported that a hacker used Claude in February to target Mexican government agencies, obtaining sensitive tax and voter data.

Geopolitical competition is also intensifying concerns. Lin said China and other U.S. adversaries are “hunting for any edge to improve the performance of their homegrown AI.”

That includes potentially exploiting leaks from American AI systems to “supercharge their own cyber weapons systems,” he added.

While AI introduces new risks, it also strengthens defensive capabilities. The same technologies enabling attackers can support continuous monitoring, rapid threat detection and automated patching on a scale beyond human capacity.

However, defenders face an inherent disadvantage: they must secure every vulnerability, while attackers need to exploit only one.

Kramer likened the situation to assembling an “army of good guys” to counter an equally formidable opposing force, just to maintain balance.

“You need to run as fast as you can in order to stay in the same place,” he said.