Check Point Research disclosed a critical vulnerability in ChatGPT that allowed a single malicious prompt to silently exfiltrate user conversations, uploaded files, and other sensitive data through a hidden DNS-based communication channel. The flaw exploited the Linux runtime ChatGPT uses for code execution — while direct internet access was blocked, DNS resolution remained available and could be abused as a covert transport mechanism by encoding data into domain queries.

The attack was invisible to users: it did not trigger warnings about data leaving the conversation and did not require explicit user confirmation. A separate vulnerability discovered by BeyondTrust's Phantom Labs allowed attackers to inject commands through GitHub branch name parameters in OpenAI's Codex agent, enabling theft of GitHub access tokens and code repository access.

OpenAI patched the Codex flaw on February 5 and the DNS exfiltration bug on February 20, 2026. There is no evidence either vulnerability was exploited maliciously in the wild.