Jerusalem, 28 October, 2025 (TPS-IL) — Israeli cybersecurity company LayerX has uncovered the first security vulnerability in OpenAI’s ChatGPT Atlas browser, allowing attackers to inject malicious instructions into ChatGPT’s memory and execute remote code. The flaw could enable hackers to gain access privileges, install malware, or take control of users’ systems.
LayerX researchers found Atlas users are up to 90% more vulnerable to phishing than Chrome or Edge users, after 97 of 103 simulated attacks succeeded. CEO Or Eshed warned that ChatGPT’s memory feature “creates a new opening for attacks” and urged users to avoid suspicious links.
The vulnerability has been reported to OpenAI for review.