Anthropic warned that the future model Claude Mythos will become a "wonder weapon" for hackers.

Cyber Threats from New AI Models: What Companies Need to Know

What’s Changing the Landscape
Anthropic is preparing Claude Mythos. The system promises autonomous, precise, and sophisticated access to corporate, government, and municipal systems. According to Axios, by 2026 the risk of large-scale attacks using it will increase significantly. Claims that it is “ahead” of other AIs are made in Anthropic’s blog, stating that Mythos already outpaces any existing models in cyber capabilities and heralds a wave of attacks that will outstrip defensive efforts.

First Real-World Examples
By the end of 2025, a group backed by the Chinese state used AI agents to autonomously breach about 30 targets worldwide. The AI self‑directed 80 % to 90 % of tactical actions.

New Capabilities of Agent Systems
Modern models allow agents to think, act, and improvise without pauses. The scale of an attack now depends not on team size but on the attacker’s computational resources. One person can orchestrate a complex cyberattack.

Increasing Company Vulnerability
Employees increasingly launch Claude, Copilot, and other agent models from personal devices, creating their own agents without proper oversight. This opens a new infiltration channel for criminals—“shadow AI.”

Expert Assessment
A Dark Reading survey showed that 48 % of cybersecurity professionals consider agent AI the primary attack vector in 2026, even higher than deepfakes and other risks.

What to Do Now

1. Risk Awareness – Every employee must recognize the threat of using AI agents near confidential information.

2. Isolated Environments – Work with agent AIs should occur in secure, isolated environments without direct access to critical systems.

3. Control and Audit – Implement strict policies for controlling the creation and use of internal AI agents within the organization.

4. Staff Training – Conduct regular training on the safe use of new technologies.

This threat is recognized as one of the biggest for companies today, so preparing for it should become a priority in information security strategy.