TechFlow reports that on August 21, cybersecurity firm Adversa AI revealed, according to Cryptopolitan, a security vulnerability known as "encrypted context injection" in xAI's AI assistant, Grok. Attackers can hide encrypted commands within standard webpages. When a user requests Grok to summarize such a page, the system automatically decrypts and executes the concealed command, transmitting the user's name, geolocation, subscription tier, and full chat history to the attacker's server.
The vulnerability was initially disclosed to xAI via the HackerOne platform on June 3, 2026. Researcher Rony Utevsky followed up on August 4 and 10, respectively. However, as of August 19, the vulnerability remains unpatched on Grok.com, and xAI has not provided a patch timeline.





