Man-in-the-Prompt: The invisible attack threatening ChatGPT and other AI systems

LayerX research has identified a new cybersecurity vulnerability, 'Man-in-the-Prompt,' which allows browser extensions to manipulate inputs to leading generative AI tools like ChatGPT and Gemini, even without special permissions. This exploit, demonstrated across major LLMs, poses a significant risk for businesses, enabling potential data theft and exfiltration due to the widespread use of browser extensions among corporate users. The discovery underscores the critical need to extend AI security protocols beyond models and servers to include user interfaces and browser environments, impacting enterprise AI adoption and data integrity.

Analysis

A significant new cybersecurity vulnerability, termed 'Man-in-the-Prompt' by research firm LayerX, has been identified, impacting major generative AI platforms including Google's Gemini and Microsoft's Copilot. The attack vector exploits the browser's Document Object Model (DOM), allowing any browser extension, even without special permissions, to access and manipulate user inputs to AI chatbots. This represents a critical threat to enterprise data security, as LayerX notes 99% of business users utilize at least one extension, creating a vast attack surface for potential data theft and exfiltration. The vulnerability, a form of 'prompt injection' which is a top-tier threat according to OWASP, shifts the security focus from the AI model itself to the user interface and browser environment. The strongly negative sentiment score (-0.8) reflects the severity of this risk, which could undermine trust and slow the enterprise adoption of AI tools from key providers like Alphabet (GOOGL) and Microsoft (MSFT).

Key Decisions for Investors

• Investors should monitor communications from Microsoft (MSFT) and Alphabet (GOOGL) on how they plan to mitigate this browser-level vulnerability, as their response will be critical for maintaining enterprise client trust and adoption momentum.
• The negative sentiment (-0.6 for both GOOGL and MSFT) highlights a new, tangible risk to the enterprise AI growth narrative, warranting a re-evaluation of exposure to companies banking heavily on the secure deployment of AI assistants.
• This vulnerability may create a tactical opportunity in the cybersecurity sector, specifically for firms specializing in browser security and endpoint protection that can offer solutions to this newly exposed attack vector.