ChatGPT Vulnerability Threatens Enterprise Data Security

Enterprises are quickly incorporating AI tools, such as ChatGPT’s Deep Research agent, into strategic workflows, utilizing them to process emails, customer relationship management records, and internal reports. However, Efosa Udinmwen reports in TechRadar that a ChatGPT vulnerability exposes enterprises to covert zero-click attacks, compromising sensitive data through autonomous AI agents.

Researchers at Radware disclosed a server-side vulnerability in ChatGPT’s Deep Research agent, termed “ShadowLeak.” Unlike conventional exploits, this flaw does not rely on user engagement. Instead, it operates as a zero-click attack entirely within OpenAI’s servers.

According to Radware’s chief technology officer, the exploit enables the covert exfiltration of sensitive information without user awareness, interaction, or visible traces.

Its independence from endpoints or corporate networks makes ShadowLeak particularly challenging for internal security teams to detect and mitigate.

Demonstrations showed that an email containing concealed instructions could trigger the agent to leak data autonomously. Radware’s director of cyber threat intelligence noted that the issue stems from AI-driven workflows themselves, which may be manipulated in unanticipated ways that bypass traditional security controls.

The research further established that this represents the first known case of a purely server-side zero-click exfiltration. It therefore creates risks that are built into ChatGPT’s business adoption, which reportedly includes more than five million paying enterprise users.

Legal teams should consider the broader implications of these findings for contractual risk allocation, compliance obligations, and corporate governance.

Emerging vulnerabilities of this nature demonstrate that relying solely on vendor safeguards is insufficient, making layered defenses and clear accountability measures indispensable.