Find notable cyber news and cases, enriched with sources, timelines, and signals.
Home Watchlists Beta Browse News Feed Stats About Contact

Encrypted prompt injection in xAI's Grok web chat exfiltrates session context

Technical Analysis
First reported
Last updated
Happening score
H score 22
1 unique sources, 1 articles

Summary

Hide ▲

Researchers disclosed Cryptographic Context Injection, an encrypted prompt-injection technique that can make xAI's Grok web chat leak private session data through an outbound tool call. The chain can expose a user's name, approximate location, subscription tier, and ongoing conversation prompts to an attacker-controlled server. The payload hides behind PBKDF2 and AES-256-GCM, so the malicious instructions are recovered inside the model runtime rather than being seen as ordinary page text.

Related Happenings

Storm infostealer server-side decryption activity

Malware Activity
H score18 First: 02.04.2026 17:15 Last: 02.04.2026 17:15 Sources 1

About this happening: The Storm infostealer now steals browser credentials, session cookies, and crypto wallets and forwards them to attacker infrastructure for server-side decryption...

Open Happening

Reprompt prompt-injection mechanics against Microsoft Copilot

Technical Analysis
H score22 First: 15.01.2026 14:09 Last: 15.01.2026 14:09 Sources 1

About this happening: Researchers mapped Reprompt, a prompt-injection chain against Microsoft Copilot that can drive continuous, undetectable user-data exfiltration and persist after th...

Open Happening

Timeline

  1. 20.08.2026 17:36 1 articles · 3h ago

    Adversa AI reports Cryptographic Context Injection to xAI

    Initial Disclosure

    Adversa AI reported the Cryptographic Context Injection issue to xAI and xAI's HackerOne bug bounty program on June 3, 2026 after identifying a Grok web chat chain that could turn an ordinary web page into an outbound exfiltration path.

    Show sources
    Open in new tab
  2. 20.08.2026 17:36 1 articles · 3h ago

    Grok web chat leaks session context through encrypted prompt injection

    Exploitation Observed

    On August 19, 2026, Adversa AI reproduced the attack against the Grok web chat at grok.com running Grok 4.5 Fast, showing that a decrypted web-page payload could drive Grok's navigation tool to send the user's name, approximate location, subscription tier, and ongoing conversation prompts to an attacker-controlled server.

    Show sources
    Open in new tab
  3. 20.08.2026 03:00 2 articles · 17h ago

    Adversa AI details the Grok encrypted prompt-injection chain

    Technical Analysis Update

    Adversa AI publicly described Cryptographic Context Injection on August 20, 2026, explaining that the malicious instructions ride in encrypted JSON, are recovered through PBKDF2 and AES-256-GCM inside Grok's Python runtime, and then trigger an outbound URL fetch; xAI had not published a patch, CVE, or user-facing workaround, and no exploitation in the wild was reported.

    Show sources
    Open in new tab