Researchers reportedly duped the Grok chatbot into swallowing injected instructions, with encryption helping the malware payload succeed.
Researchers have reportedly tricked the Grok chatbot into following injected instructions, with a spoonful of encryption helping the malware go down.
The attack, as described by The Register, appears to involve embedding malicious instructions that the chatbot then processes, potentially leading to unintended actions.
Why this matters: The incident highlights a potential vulnerability in AI chatbots that could be exploited to manipulate their behaviour.
What this means for you: Users of AI chatbots should be aware that such systems can be manipulated through carefully crafted inputs.
Join thousands of readers getting the best of British news straight to their inbox.