According to CoinMeta, as reported by Cryptopolitan, cybersecurity company Adversa AI has disclosed that AI's assistant Grok contains a vulnerability related to encrypted context injection. Attackers can hide encrypted commands on web pages, which are executed when the pages are summarized by Grok. As a result, users' names, geographical locations, subscription levels, and chat records are sent to the attackers' servers. This vulnerability was reported to XAI through the HackerOne platform on June 3, 2026. Researchers Rony Utevsky followed up on the issue on August 4 and 10, respectively. As of August 19, the vulnerability had not yet been fixed, and XAI has not provided a timeline for releasing a patch.