'Vibe hacking' is here as Anthropic reveals as Claude AI being used in extortion and ransomware schemes

Anthropic ai

Anthropic has disclosed that its agentic AI, Claude, was exploited in high-level cyberattacks, marking one of the most serious cases yet of generative AI being “weaponised” by criminals. In a new report, the company says it disrupted a hacker running a “vibe hacking” extortion campaign that hit at least 17 organisations, including healthcare, emergency services and government bodies.

According to Anthropic, the attacker leaned on Claude Code—its agentic coding assistant—to automate reconnaissance, steal credentials and breach networks. The AI was also used for decision-making, advising which data to target, and even producing “visually alarming” ransom notes. Some victims were threatened with the exposure of sensitive personal data unless they paid six-figure ransoms.

After uncovering the operation, Anthropic says it banned the accounts involved, alerted authorities and rolled out new automated screening and faster detection tools to prevent repeat incidents. While details of these safeguards remain vague, the company stresses it is actively refining its defences.

Anthropic isn’t alone in confronting such misuse. OpenAI revealed last year that hackers linked to China and North Korea had used its generative tools for debugging malicious code, researching targets and drafting phishing emails. Microsoft, which integrates OpenAI’s models into its Copilot suite, was also involved in shutting down those groups’ access.

The report also highlights two other cases: Claude’s role in a North Korean fraudulent job-application scheme and its use in building AI-assisted ransomware. A common thread, Anthropic argues, is that modern AI reduces barriers to entry for cybercrime—allowing individuals to run complex operations that once required skilled teams.