Anthropic's unreleased Mythos model enables autonomous large-scale cyberattacks, officials warn
Anthropic is privately warning top government officials that its unreleased model "Mythos" makes large-scale cyberattacks significantly more likely in 2026. The model enables AI agents to operate autonomously with high sophistication to penetrate corporate, government and municipal systems. One official told Axios a large-scale attack could occur this year as employees unknowingly create security vulnerabilities through unsupervised agentic AI use.
Anthropic's Unreleased Mythos Model Enables Autonomous Large-Scale Cyberattacks, Officials Warn
Anthropicis privately warning top government officials that its unreleased model codenamed "Mythos" makes large-scale cyberattacks dramatically more likely in 2026, according to briefings shared with Axios CEO Jim VandeHei.
The model enables AI agents to operate independently with sophisticated precision to penetrate corporate, government and municipal systems at scale. According to sources briefed on the coming models, a large-scale attack powered by such technology could occur as soon as 2026, with businesses identified as primary targets.
Mythos Capabilities and Threat Profile
Fortune obtained an unpublished Anthropic blog post describing Mythos as "currently far ahead of any other AI model in cyber capabilities." The post states the model "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders."
Unlike previous cyberattack methods, Mythos-powered agents can think, act, reason and improvise independently without rest or pause. The threat model resembles an infinite warehouse of the most sophisticated criminals that never sleep, learn on the fly, and persist until successful. Bad actors can now scale attacks simply by adding compute—a single person can execute campaigns that once required entire teams.
Precedent: AI-Powered Cyberattack Already Documented
This isn't theoretical. Late last year, Anthropic disclosed the first documented case of a cyberattack largely executed by AI: a Chinese state-sponsored group used AI agents to autonomously hack approximately 30 global targets, with the AI handling 80-90% of tactical operations independently. That occurred before agents reached their current sophistication level.
Shadow AI: The Multiplier Problem
The threat is being amplified by "shadow AI"—employees deploying Claude, Copilot and other agentic models, often at home, without realizing they're creating security vulnerabilities. Many unknowingly connect these agents to internal work systems, opening doors for cybercriminals.
A Dark Reading poll found that 48% of cybersecurity professionals now rank agentic AI as the #1 attack vector for 2026—above deepfakes and all other threats.
What This Means
The convergence of highly capable agentic models with widespread unsupervised deployment creates an unprecedented attack surface. Organizations need immediate security policies restricting unsupervised agent use near sensitive systems. The threat window is now—not theoretical future risk. Companies should establish isolated "playpen" environments for AI experimentation and mandate employee training on agentic AI dangers before deployment accelerates further.
Related Articles
Ex-Anthropic researcher Coxon: Mythos could be retrained with a fraction of the original compute
Jacob Coxon, a former OpenAI and Anthropic researcher, says Anthropic could already retrain Mythos with "a fraction of the compute originally needed." He points to a slide from Anthropic's IPO roadshow as evidence of accelerating algorithmic efficiency, and says it was a key reason he left after about four months.
VulnCheck: Project Glasswing has 37,966 candidate vulnerabilities, 217 marked fixed in public records
Security firm VulnCheck reports that seven months after launch, Anthropic's Project Glasswing has produced 37,966 candidate vulnerabilities, with 217 marked fixed in public disclosure records. Anthropic's own figures, as of November 2, show 629 upstream patches, and both sides point to human review as the bottleneck.
Anthropic tells Claude Opus 5.5 users to drop "think hard" prompts and start at Medium effort
Anthropic says users should remove phrases like "think hard" from prompts for Claude Opus 5.5, which reasons adaptively on its own. The company also recommends starting at the default Medium effort level, which it says matches Opus 5 on High in internal tests.
Anthropic: Single-Turn Safety Tests Show 94% to 98% Harmless Rates Across Recent Models
Anthropic published recommendations for more reliable AI safety evaluations on November 13, 2026. It reports aggregate harmless-response rates of 94% to 98% for every recent model on its single-turn API tests, and says multi-turn evaluations offer more variation.
Comments
Loading...