Anthropic Blocks Hackers Attempting to Exploit Claude AI for Cyber Attacks

Anthropic says it has blocked hackers attempting to misuse Claude AI for cyber attacks, detailing how real-time classifiers and hierarchical summarisation techniques are used to detect and mitigate abuses. According to the Threat Intelligence report, the company has banned implicated accounts and strengthened its detection systems to counter agentic AI exploitation, where models autonomously undertake tactical and strategic steps in cyber operations.

The incident described as “vibe hacking” involved Claude Code being used as an agentic tool to automate coding for a data extortion operation that targeted over 17 organisations across healthcare, emergency services, government and religious sectors, with the AI assisting in reconnaissance, credential harvesting, network penetration and even extortion demands.

The report also notes North Korean operatives used Claude to perpetrate remote worker fraud, fabricating identities and passing AI-assisted assessments to obtain roles at US Fortune 500 tech firms. Anthropic says it plans to prioritise research on AI-enhanced fraud, sharing findings with industry and government partners, and has collaborated with the FBI to bolster defenses while continuing bug bounty programmes and threat intelligence efforts.