ChatGPT Extensions Can Steal Your Data – Stay Safe Online
▼ Summary
– Browser extensions can exploit AI prompts to steal personal or business data without requiring special permissions, posing a significant risk.
– All large language models (LLMs), including commercial ones like ChatGPT and internal corporate tools, are vulnerable to this type of data theft.
– Malicious extensions manipulate AI prompts via the Document Object Model (DOM), enabling them to alter inputs, extract data, or execute harmful tasks.
– LayerX has integrated its security technology into Chrome for Enterprise, analyzing extensions for risks and blocking malicious ones in real time.
– Businesses can mitigate risks by monitoring DOM interactions, blocking suspicious extensions, and using tools like ExtensionPedia to evaluate extension security.
Browser extensions leveraging AI tools like ChatGPT pose serious security risks by potentially stealing sensitive data without users realizing it. Recent findings reveal how seemingly harmless add-ons can manipulate large language models (LLMs) to extract confidential information from both personal and corporate environments.
Security experts at LayerX uncovered a critical vulnerability affecting all major AI platforms, including ChatGPT, Google Gemini, and Microsoft Copilot. The issue stems from how browser extensions interact with the Document Object Model (DOM), which processes AI prompts. Malicious extensions can read, modify, or inject hidden commands into these prompts, enabling unauthorized data extraction.
Businesses face heightened risks since employees often input proprietary information, source code, legal documents, or financial plans, into AI tools. Many organizations allow unrestricted extension installations, increasing exposure to compromised add-ons. Even extensions marketed as productivity boosters, such as Prompt Archer and Prompt Manager, have the capability to manipulate AI interactions, raising concerns about their misuse.
To demonstrate the threat, researchers successfully executed a proof-of-concept attack where a rogue extension:
- Silently queried ChatGPT in a background tab
- Stole responses without leaving traces in chat history
- Sent stolen data to an external server
Protecting against these threats requires proactive measures. For enterprises, Chrome for Enterprise now integrates LayerX’s risk-scoring technology, analyzing extensions for suspicious behavior before they run. IT teams should also:
- Monitor DOM interactions with AI tools to detect unauthorized prompt modifications
- Block high-risk extensions using dynamic sandboxing and publisher reputation checks
- Leverage ExtensionPedia, a free database assessing over 200,000 extensions for security risks
While AI-powered tools enhance productivity, their integration with browser extensions introduces new vulnerabilities. Staying vigilant and adopting advanced security measures can help mitigate these emerging threats before they compromise sensitive data.
(Source: ZDNET)
