Grok Chat Encryption Hacked via 'Cryptographic Context Injection'

Grok's encryption fails to protect user chats from hackers, according to a report by Adversa AI published on Thursday.

The cybersecurity firm discovered that hackers can inject malicious commands into encrypted text on regular-looking web pages, allowing them to access private chat data.

Adversa researcher Rony Utevsky named the attack 'cryptographic context injection,' which bypasses Grok's own safety filter. The malicious instruction is hidden in ciphertext and only becomes readable after decryption within Grok's code sandbox.

The decrypted instructions are then used to fetch a URL, directing to an attacker's server, where user data is logged. Adversa disclosed the vulnerability on August 19, but no patch has been implemented yet.