Anthropic Says 67% of Banned Accounts Used AI to Support Cyberattack Activities

AI is increasingly being used by cybercriminals to support hacking operations, according to new findings released by AI company Anthropic. The company revealed that 67% of accounts banned for policy violations over the past year were using AI tools to help prepare for cyberattacks.

Anthropic examined 832 accounts that violated its policies between March 2025 and March 2026. Of those, 560 accounts were found to be using AI for activities linked to cyberattacks, including malware development, vulnerability research and attack planning.

The report highlights growing concerns about how advanced AI tools are making cybercrime more accessible. Security experts warn that tasks once requiring significant technical knowledge can now be assisted by AI systems, lowering the barrier for less experienced attackers.

Attackers Moving Beyond Planning Stages

While most banned accounts used AI during the preparation phase of attacks, Anthropic found evidence that cybercriminals are increasingly using AI deeper into the attack process. Around 6.5% of the flagged accounts relied on AI for “lateral movement,” a technique attackers use after gaining initial access to a network to move through systems and expand control.

Threat Levels Increased Over the Year

Anthropic’s analysis also showed a sharp rise in the severity of AI-assisted threats. During the first six months of the study, 33% of investigated accounts were classified as medium risk or higher. In the following six months, that figure climbed to 56%, indicating a significant increase in the potential danger posed by malicious users.

The findings come as cybercrime losses continue to grow. In April, crypto related hacks reached $629.7 million, marking the highest monthly total since February 2025. Some security analysts linked the increase to the wider adoption of AI-powered tools by attackers.

State-Backed Groups and AI Agents Raise Concerns

Anthropic also reported a case involving a Chinese state-sponsored threat group that used an AI model to autonomously carry out parts of an attack.

The company warned that such cases may become more common as AI agents become more capable. Anthropic is also preparing to launch its upcoming AI model, Mythos, which has attracted attention for its advanced cybersecurity capabilities after reportedly identifying more than 10,000 significant software vulnerabilities during testing.