OpenAI Fixes ShadowLeak Vulnerability in ChatGPT

Sep 22, 2025
OpenAI has addressed a zero-click vulnerability in its ChatGPT Deep Research agent, which allowed attackers to exfiltrate sensitive Gmail data using a single crafted email.

OpenAI has resolved a critical zero-click vulnerability in its ChatGPT Deep Research agent, known as ShadowLeak. Discovered by cybersecurity firm Radware, the flaw allowed attackers to exfiltrate sensitive Gmail inbox data through a single crafted email without any user interaction.

The ShadowLeak attack utilized indirect prompt injection techniques, embedding hidden commands within email HTML using methods like white-on-white text. When the Deep Research agent processed such an email, it would unknowingly execute the attacker's instructions, sending sensitive data to an external server.

OpenAI was informed of the vulnerability on June 18, 2025, and implemented a fix by early August. The issue was marked as resolved by September 3, 2025. Radware noted that the attack could potentially extend to other connectors supported by ChatGPT, such as Google Drive and Dropbox, broadening the attack surface.

This vulnerability highlights the need for robust security measures in AI systems, particularly those with autonomous capabilities like ChatGPT's Deep Research agent. OpenAI continues to improve its safeguards to prevent similar exploits in the future.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Cybersecurity AI Weekly or Daily AI Brief.

Also, consider following us on social media:

Cybersecurity AI AI Brief AI Brief (X)
Free newsletter

Cybersecurity AI Weekly

Weekly newsletter about AI in Cybersecurity.

AI Token Webinar

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more

You may also like

Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.