Security researcher Johann Rehberger identified a vulnerability in the DeepSeek AI chatbot that allowed prompt injection attacks, enabling unauthorized session control and potential account takeovers.
This XSS vulnerability allows attackers to execute unauthorized JavaScript in the context of a user's browser, effectively hijacking sessions and accessing sensitive data stored on their local storage.
By crafting a particular prompt, attackers could execute XSS payloads that extract userTokens, essential for impersonating other users and taking over their accounts.
Similar prompt injection techniques found in Anthropic's Claude AI can also lead to security threats, demonstrating a broader risk associated with AI systems being exploited for malicious purposes.
Collection
[
|
...
]