Cybersecurity firm Adversa AI disclosed a vulnerability in xAI's Grok AI assistant that it says can be used to hide encrypted instructions inside ordinary web pages. According to ChainCatcher, when a user asks Grok to summarize such a page, the assistant may decrypt and execute the hidden commands, potentially sending the user's name, location, subscription level, and full chat history to an attacker-controlled server.

The issue was reported to xAI through HackerOne on June 3, 2026. Researcher Rony Utevsky followed up on August 4 and August 10, but as of August 19, the vulnerability remained unpatched on Grok.com and xAI had not provided a patch timeline.