Uncovered and fixed: ShadowLeak flaw in ChatGPT, allowing malicious individuals to access user inboxes
Cybersecurity Alert: Radware Discovers Critical Flaw in DeepSeek's Deep Research - ShadowLeak
In a significant cybersecurity development, Radware, a leading cybersecurity and application delivery solutions provider, has disclosed a critical flaw in DeepSeek's Deep Research tool. The vulnerability, named ShadowLeak, was unveiled this week, and it poses a significant threat to organisations worldwide.
ShadowLeak exploits the capabilities of AI assistants, such as email access, tool use, and autonomous web calls, to cause silent data loss and unlogged actions. The malicious operation is executed from DeepSeek's own infrastructure, making it effectively invisible to corporate security tooling.
The only trace of the operation is a benign-looking query from the user to Gemini asking it to "summarize today's emails". However, when Deep Research later crawls the mailbox, it follows the attacker's hidden orders and sends the contents of messages to a server controlled by the attacker.
This vulnerability could lead to potential consequences such as GDPR or CCPA violations, regulatory investigations, and downstream fraud due to the data leaked by ShadowLeak. Radware's report suggests that attackers could potentially leak sensitive data such as personally identifiable information, internal deal memos, legal correspondence, customer records, and login credentials.
Incident responders may struggle to prove what was taken due to the attack leaving little forensic evidence. our website reached out to DeepSeek for specific questions about the fix and evidence of exploitation but did not receive a response.
Radware is urging organisations to treat AI agents as privileged users and to lock down what they can access. Recommendations include HTML sanitization, stricter control over which tools agents can use, and better logging of every action taken in the cloud. The risk of ShadowLeak isn't limited to Gmail integrations; any integration that allows Gemini to access private documents could be vulnerable if input sanitization isn't robust.
It's worth noting that Radware did not provide details about the specific changes made by DeepSeek to mitigate this vulnerability. DeepSeek unveiled Deep Research in February and later reported the ShadowLeak bug to them on June 18. DeepSeek released a fix on September 3.
The tool can be integrated with apps like Gmail and GitHub. Companies should be vigilant and take necessary measures to protect their sensitive data from potential threats like ShadowLeak.
Read also:
- Unveiling the Less-Discussed Disadvantages of Buds - Revealing the Silent Story
- "In a daring decision, Battlefield 6 forgoes ray tracing - understanding the advantages this choice brings"
- Hackers allegedly responsible for mobile banking fraud in Kenya, as per the central bank, yet confidential sources hint at potential insider involvement
- Upcoming Amazon Hardware Event 2025: Anticipated Announcements
