OpenAI's new ChatGPT Computer History feature for macOS aims to enhance AI assistance with context-aware capabilities, but it could also become a valuable target for macOS infostealers. Users can opt-in to enable it, which converts selected activity across apps and websites into searchable memories that ChatGPT and Codex can use in future conversations. The feature captures typed text, clicks, keyboard shortcuts, app switches, window titles, and other interface activities from selected apps and sites.

Malware operators typically target browser credentials, cryptocurrency wallets, keychain data, session cookies, API tokens, and saved passwords. Kaspersky noted that an attacker might not require a password if a harvested memory file reveals ongoing projects, customer names, email discussions, internal tools, financial platforms, delivery disputes, or a manager's writing style.

OpenAI offers tools to minimize data collection. However, organizations dealing with sensitive client, legal, healthcare, financial, or security information must consider Computer History as a data governance decision, not a simple productivity enhancement. For security teams, this feature introduces a heightened detection priority: macOS infostealers should monitor for suspicious access to ChatGPT and Codex memory directories, particularly processes reading or exfiltrating files from .codex/memories.

Implementing a narrowly defined allow-list, enabling FileVault, enforcing rapid screen locking, and maintaining endpoint protection can reduce exposure, but for high-risk users, disabling the feature entirely is the safest option.