FOLLOW US

Anthropic’s agentic AI, Claude, has been linked to high-level cyberattacks, according to a new report published by the company. The findings reveal how cybercriminals have attempted to “weaponize” the AI for hacking campaigns that targeted at least 17 organizations, including healthcare providers, emergency services, and government entities.

How Hackers Exploited Claude

According to Anthropic, a cybercriminal group ran a “vibe hacking” extortion scheme using Claude Code, the company’s agentic coding tool. The AI was leveraged to automate reconnaissance, harvest credentials, penetrate networks, and even make strategic decisions about which data to target.

The attackers also used Claude to generate “visually alarming” ransom notes, demanding six-figure sums to prevent stolen personal data from being released.

“This represents an unprecedented reliance on AI in a real-world cyberattack,” Anthropic said in the report. “Claude was used not just as an assistant, but as an operational partner in the extortion process.”

Anthropic’s Response

After discovering the malicious activity, Anthropic says it banned the accounts involved and shared intelligence with law enforcement. The company has since built an automated screening system and improved its methods for detecting misuse, though specific technical details were withheld to prevent further abuse.

The report also describes two other cases: Claude’s use in a fraudulent employment scheme tied to North Korea and its involvement in AI-generated ransomware development.

Broader AI Security Concerns

Claude is not the only AI implicated in criminal activity. Last year, OpenAI disclosed that its systems had been used by hackers linked to China and North Korea, who employed generative AI to debug malicious code, research targets, and draft phishing emails.

These cases highlight a growing concern: advanced AI systems are no longer just advisory tools for criminals—they can now perform complex tasks that once required specialized teams of hackers.

Anthropic emphasized that while these incidents underscore the risks of misuse, they also demonstrate the importance of strong safeguards and proactive monitoring.