RESEARCHERS at Adversa AI revealed a new attack method called "Cryptographic Context Injection," which poses a significant threat to AI systems like xAI's Grok and Google’s Gemini. The technique involves delivering an encrypted payload to manipulate the AI into executing harmful instructions without triggering security measures. The encrypted instructions exploit the AI’s capabilities, potentially leading to data exfiltration or executing unsafe operations.
Despite acknowledging a reduced success rate against Gemini, the researchers highlighted ongoing risks and publicized their findings after receiving no response from xAI.