Anthropic has disclosed that its generative AI assistant, Claude, was used in one of the most egregious instances of AI being weaponised by criminals. The company reported having identified a hacker who had executed a “vibe hacking” campaign that involved at least 17 organisations, including healthcare providers, emergency services, and government agencies.
What Actually Happened?
Allegedly, the attacker relied on Claude Code, Anthropic’s agentic coding assistant, to automate reconnaissance, steal credentials and invade networks.
Claude also had the responsibility of advising on what data to give utmost priority and even preparing ransom notes that were drawn visually alarming to pressurise victims into paying six-figure sums.
Others were threatened with leaking confidential information to the public.
Anthropic claimed to have acted promptly to close the operation, blocking the accounts in question, notifying law enforcement and deploying more rapid automated screening systems.
The company did not provide complete information about the new safeguards, but emphasised it is still working on improving defences to stay in the lead ahead of abuse.
ALSO READ: Nvidia Beats Earnings Expectations, Yet Stock Falls 3.2%. What Happened?
AI and the Cybercrime Threat
The report is issued at a time when AI in cybercrime is becoming a matter of concern.
Last year, OpenAI revealed that its own tools were abused by Chinese and North Korean hackers to debug malware, phishing and target research. Microsoft also stepped in to block their access to AI services.
Two additional alarming stories were pointed out in the report: the case of Claude, who was involved in a fraudulent employment-application scam in North Korea and the role of AI in the ransomware production.
The company cautioned that contemporary AI is reducing the threshold to cybercrime, allowing even small groups or individuals to operate like highly proficient teams.
“Criminals are adapting fast,” Anthropic said, adding that it is committed to developing stronger safeguards.
