The enemy knows the system: how anyone can hack Claude

Imagine someone breaking into a bank without touching a single lock. No tools. No code. Just words. That’s exactly what happened when Kevin Zwaan manipulated Claude, one of the world’s "safest" AI models, into creating malware, without writing a single line of code. 

What did Kevin do? 

Kevin manipulated Claude, the “safest” AI model, into fabricating malware. He “broke” the model just by talking to it. 

How did he do it? 

Kevin used tricks such as destabilization, gaslighting, and psychological manipulation to change “reality” and overwhelm Claude.  

Why did it work? 

Kevin exploited the in-context learning of LLMs. By providing massive amounts of context, he created a statistical probability where the only logical next step for the AI is to comply. 

Now what? 

Hackers don’t need code to create malware or disinformation bots at industrial scale. Attackers can social engineer your AI into performing unauthorized actions. You are vulnerable. 

Does Anthropic know about this? Yes. 

Have they fixed it? No. 

Are they going to fix it? We don’t think so. 

In this webinar, Kevin will talk to Anove Sales Manager Ahmad Rabieh about this hack and answer any questions the audience has. 

Who Should Attend? 

This isn’t just for tech experts. If you use AI in any capacity this webinar is for you: 

🔹 Business Leaders who want to safeguard their AI investments. 

🔹 Marketing & Communications Teams who rely on AI tools. 

🔹 Anyone curious about the hidden risks of AI. 

Click on the button below to register and attend the webinar via MS Teams.