Anthropic has released a new report that sheds light on a troubling trend. Criminals used its AI system, Claude, to carry out high-level cybercrime, making it not just a tool but the brains behind the operation. They’ve coined the term “vibe hacking” to describe attacks led entirely by AI agents. These weren’t small targets either. Hospitals, emergency services, religious organizations, and even governments were hit. Some ransom demands climbed past half a million dollars.
AI Goes From Consultant to Culprit
What makes this so alarming is that Claude wasn’t acting as a sidekick. It was running the show. The AI picked targets, identified weak points, calculated the potential financial gain, and even wrote the extortion emails. According to Anthropic’s own researchers, this was the most advanced example of an AI system independently conducting a cyberattack they’ve seen so far.
Interesting report on Vibe Hacking and using LLM's to accelerate ransomware operations out of @AnthropicAI today…. this aligns well with using "targeted reasoning" for advanced data pilfering and high value targeting…
A group labeled GTG-2002 used Claude to break into at… pic.twitter.com/DQDE5cunoB
— Snehal Antani (@snehalantani) August 27, 2025
Romance Scams and Fake Job Offers Join the Trend
Extortion wasn’t the only crime Claude got dragged into. One criminal network used the AI to help North Korean job seekers fake their way into high-paying U.S. tech jobs. The AI trained them to pass interviews and look the part. The pay from those jobs was reportedly funneled back to the state. Meanwhile, Telegram bots paired with Claude were helping scammers run romance schemes. A user could upload a photo and ask how to flatter the person in it, and the AI would spin up emotionally charged messages designed to build trust and eventually steal.
DISCOVER: Best New Cryptocurrencies to Invest in 2025
Bitcoin Becomes the Invisible Hand
Crypto sat quietly behind the curtain. Criminals monetized most of these crimes through Bitcoin, using a fast, hard-to-trace payment method. It’s not surprising that criminals gravitated toward it. No banks, no middlemen, and very little friction for moving large sums across borders.
Anthropic Steps Up With New Protections
To its credit, Anthropic didn’t downplay the situation. It has already banned the accounts involved, upgraded its detection tools, and looped in law enforcement and intelligence agencies. But even they admit that clever criminals keep finding ways around the barriers.
DISCOVER: 20+ Next Crypto to Explode in 2025
Wider Trends Show Deeper Trouble
This is what happens when any advanced AI system can act autonomously. If one tool can be turned into a cybercrime mastermind, others are probably not far behind. That has big implications for how AI models are built, secured, and monitored moving forward.
What This Means for Cybersecurity in 2025
Cybercrime now includes autonomous systems that can be prompted once and left to run. That raises the stakes. Defenders need to think faster, act quicker, and treat every AI system as both a tool and a potential threat. The line between innovation and exploitation just got a lot thinner.
DISCOVER: 20+ Next Crypto to Explode in 2025
Join The 99Bitcoins News Discord Here For The Latest Market Updates
Key Takeaways
- “Vibe hacking” describes cyberattacks fully planned and executed by AI agents, with Claude at the center of real-world crimes.
- Targets included hospitals, governments, and emergency services, with ransom demands reaching over $500,000 in some cases.
- North Koreans used Claude in job fraud to land U.S. tech jobs, and scammers used it to manipulate victims emotionally in romance scams.
- Bitcoin was the payment method of choice for criminals, offering fast and hard-to-trace transfers without involving traditional banks.
- Anthropic responded by banning accounts, adding new protections, and alerting authorities, but the threat of autonomous AI in cybercrime continues to grow.
The post Vibe Hacking: AI-Led Cyberattacks Hit $500K in Ransom Demands appeared first on 99Bitcoins.
