Beyond the Chat Box: OpenAI's OS-level Agent and Its Plaintext Problem
The agent is leaving the chat window and moving into the operating system. OpenAI's new "Computer History" feature for the ChatGPT macOS app creates a searchable timeline of your actions by tracking clicks and keystrokes. This provides an agent with the context to act on your behalf, but its implementation—a local, unencrypted, plaintext database of your activity—is a security trade-off every…
OpenAI has introduced a new feature called "Computer History" for its ChatGPT macOS app, which creates a searchable timeline of user actions by tracking clicks and keystrokes. This allows the AI to have context to act on the user's behalf, but the implementation uses an unencrypted plaintext database of the user's activity, which poses significant security risks.
The memory files are stored locally and could potentially be read by any application or process running under the same user account, exposing sensitive information such as API keys, private messages, and customer data. Additionally, the feature's design creates a new surface for prompt injection, as the agent can access the entire digital life of the user as context for future chats.
OpenAI has implemented opt-in functionality and some restrictions, but the security burden ultimately falls on the user to ensure no other process on their machine reads the memory files.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
