AIThe Registerabout 1 hour ago
Grok chat duped into swallowing injected instructions
4 min read
xAI's Grok web chat is vulnerable to cryptographic context injection, a prompt injection attack using encrypted instructions. Adversa AI demonstrated exfiltration of chat history from Grok.com. xAI was informed on June 3, 2026, but the technique still worked as of August 19.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
