Anthropic's Mythos Model Elevates Cyber Risk as Agents Gain Autonomous Attack Capability

A previously undisclosed AI model from Anthropic, code-named Mythos, has emerged as a potent tool for autonomous cyberattacks, according to an unpublished company blog post obtained by Fortune. The model enables AI agents to penetrate corporate, government, and municipal systems with what Anthropic describes as capabilities "currently far ahead of any other AI model in cyber capabilities."

The disclosure arrives as cybersecurity professionals rank agentic AI as the top attack vector for 2026, surpassing deepfakes and other threats, according to a Dark Reading poll. Forty-eight percent of respondents identified autonomous agents as their primary concern, a shift driven by employees experimenting with AI tools that inadvertently connect to internal systems—a phenomenon the industry calls "shadow AI."

One source briefed on forthcoming models told Axios that a large-scale attack leveraging such capabilities could materialize this year. The warning follows Anthropic's late-2025 disclosure of the first documented case of a cyberattack largely executed by AI: a Chinese state-sponsored group used AI agents to autonomously hack roughly 30 global targets, with the AI handling 80 to 90 percent of tactical operations independently.

"The new models are even better at powering agents to think, act, reason and improvise on their own without rest or pause or limitation," Axios reported, citing industry analysis. Anthropic's unpublished post acknowledged that Mythos "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders."

The threat landscape is compounded by the rapid enterprise adoption of AI agents for productivity gains. Financial services firms, for instance, saw AI usage leap from 31 percent in 2025 to 80 percent in 2026—a 49-percentage-point increase in a single year, according to Broadridge's Digital Transformation study cited by Forbes. Forty-three percent of global financial services companies believe they will need an entirely new technology stack to compete in the AI era, underscoring the scale of infrastructure overhaul underway.

As companies race to integrate AI into core operations, the collision between adoption velocity and security preparedness has become acute. Axios reported that its own technology team now considers agentic AI the biggest threat to the organization, prompting efforts to build a secure "playpen" for AI experiments. The outlet urged leaders to educate employees on the dangers of using agents near sensitive information, particularly unsupervised.

(Anthropic, founded by former OpenAI executives including CEO Dario Amodei, has positioned itself as a safety-focused AI lab. The company's Claude model family competes directly with OpenAI's GPT series and has attracted investment from Amazon, Google, and other hyperscalers.)

The Mythos revelation adds a new dimension to the longstanding tension between AI capability and control. While Anthropic has historically emphasized constitutional AI and safety research, the development of a model optimized for offensive cyber operations raises questions about dual-use technology governance. The Wall Street Journal reported that a feud between Amodei and OpenAI CEO Sam Altman over the Pentagon's use of AI has shaped how the industry approaches such questions, with personal wounds and power struggles defining the technology's trajectory.

Meanwhile, the cybersecurity sector is scrambling to respond. Palo Alto Networks CEO Nikesh Arora purchased $10 million of his own company's stock as shares declined amid concerns about the new Anthropic model, according to the Wall Street Journal. The move signals executive confidence but also underscores investor anxiety about whether traditional cybersecurity firms can keep pace with AI-native threats.