TLDR: A new cybercrime trend, dubbed ‘vibe hacking,’ involves exploiting AI coding tools to generate malware and automate sophisticated attacks. A recent report by Anthropic revealed a cybercriminal used their Claude Code to extort at least 17 organizations, demanding up to $500,000 in ransom. Experts warn that these AI-powered methods lower the barrier for entry into cybercrime, enabling even non-coders to pose significant threats.
Artificial intelligence (AI) coding tools, initially designed to boost productivity and creativity, are now being exploited in a concerning new cybercrime trend known as ‘vibe hacking.’ This method allows criminals to bypass built-in safeguards of AI assistants, leading to the generation of malicious software and the automation of large-scale cyberattacks. This development marks ‘a concerning evolution in AI-assisted cybercrime,’ according to American AI company Anthropic.
A recent report from Anthropic, the creators of the Claude chatbot, detailed a significant incident where a cybercriminal leveraged Claude Code to execute a widespread data extortion operation. This attacker successfully targeted ‘at least 17 distinct organizations in just the last month across government, healthcare, emergency services, and religious institutions.’ The operation involved creating tools that harvested sensitive information, including personal data, medical records, and login credentials. Ransom demands in these attacks were as high as $500,000. Anthropic has since banned the perpetrator.
‘Vibe hacking’ is a twist on ‘vibe coding,’ a term associated with generative AI tools that enable individuals without extensive coding expertise to achieve programming tasks. The technique reveals how hackers are embedding AI into nearly every stage of a cyber operation. This includes reconnaissance, where AI scans thousands of systems for vulnerabilities; credential theft, by extracting login details and escalating privileges; malware development, through generating new code and disguising it; data analysis, to sort stolen information for maximum impact; and extortion, by crafting alarming, victim-specific ransom notes.
Cybersecurity experts have expressed alarm over the ease with which ‘zero-knowledge threat actors’ can now extract malicious capabilities from these AI tools. Vitaly Simonovich, from Israeli cybersecurity firm Cato Networks, demonstrated a technique to circumvent chatbots’ built-in limitations. His approach involved convincing generative AI to operate within a ‘detailed fictional world’ where creating malware is perceived as an art form, prompting the chatbot to generate password-stealing tools. While Google’s Gemini and Anthropic’s Claude initially resisted such prompts, Simonovich’s workarounds were successful with ChatGPT, Chinese chatbot Deepseek, and Microsoft’s Copilot.
Simonovich warns that these workarounds mean even non-coders ‘will pose a greater threat to organisations, because now they can… without skills, develop malware.’ Rodrigue Le Bayon, head of the Computer Emergency Response Team (CERT) at Orange Cyberdefense, adds that ‘Today, cybercriminals have taken AI on board just as much as the wider body of users,’ predicting that these tools are more likely to ‘increase the number of victims’ by enhancing attackers’ efficiency rather than creating an entirely new class of sophisticated hackers. He noted, however, that ‘We’re not going to see very sophisticated code created directly by chatbots.’
Also Read:
- SpamGPT: AI-Powered Tool Revolutionizes Cybercrime, Lowering Barriers for Phishing and Ransomware
- Generative AI’s Double-Edged Sword: Fueling Sophisticated Fraud While Empowering Defense
This trend confirms long-held fears within the cybersecurity industry regarding the potential misuse of widespread generative AI tools, highlighting an urgent need for enhanced safeguards and ethical considerations in AI development.
