Full Report
'Computer History' records clicks and typing to build ChatGPT memories
Analysis Summary
# Industry News: OpenAI Launches 'Computer History'—A Text-Based Alternative to Visual Surveillance
## Summary
OpenAI has introduced "Computer History," an opt-in feature for macOS that records user interactions—including typing, clicks, and app usage—to build a persistent memory timeline for ChatGPT. Positioned as a less intrusive alternative to screenshot-based tracking, the tool aims to enhance automation and context-awareness while raising significant privacy and security concerns.
## Key Details
- **Date:** August 14, 2026
- **Companies Involved:** OpenAI
- **Category:** Product Update / Feature Launch
## The Story
Following the controversial industry trend of "total recall" style tracking (pioneered by Microsoft), OpenAI has released "Computer History" for its macOS desktop app. Unlike Microsoft’s Recall, which relies on periodic screenshots, OpenAI’s implementation functions as a sophisticated keylogger and event capture system.
The tool utilizes macOS accessibility APIs to monitor clicks, keyboard shortcuts, and text input across various applications and websites. This raw data is periodically summarized into "memories" and organized into a timeline. While the raw event files are deleted after 48 hours, the generated summaries are stored as persistent memory files, which OpenAI uses to inform future ChatGPT responses and identify automation opportunities. The feature is currently restricted to Pro, Business, and Enterprise tiers and is notably absent from the EEA, UK, and Switzerland due to stringent privacy regulations.
## Business Impact
### For the Companies Involved
- **OpenAI:** Increases user "stickiness" by creating a deeply personalized AI assistant that knows the user's specific workflows. It also drives revenue through increased token consumption required for summarizing activity logs.
### For Competitors
- **Microsoft/Apple/Google:** Sets a new benchmark for "agentic" AI. It shifts the battleground from "General LLMs" to "Context-Aware Personal Assistants."
- **Niche Productivity Tools:** Directly threatens specialized time-tracking and workflow-automation startups.
### For Customers
- **Enterprise:** Offers potential gains in employee productivity through automated context-switching and task resumption.
- **Individual Users:** Provides a powerful tool for self-quantification and workflow automation at the cost of high-granularity data exposure.
### For the Market
- **Surveillance Normalization:** Continues the industry trend of "opt-in surveillance," where advanced utility is traded for deep behavioral monitoring.
## Technical Implications
The system relies on an interaction-event stream rather than pixel processing. By converting events into text summaries locally before sending them to OpenAI servers for memory generation, it reduces bandwidth but creates a local "honeypot" of unencrypted sensitive data accessible to other processes on the machine.
## Strategic Analysis
- **Market Positioning:** OpenAI is positioning itself as the "Operating System of Context," moving beyond the chat box and into the user’s background workflow.
- **Competitive Advantage:** Deep integration with user activity allows OpenAI to solve the "blank slate" problem of AI—where the AI knows exactly what the user was doing five minutes ago in a different app.
- **Challenges:** Significant legal hurdles in the EU/UK; massive privacy blowback; and the inherent danger of storing unencrypted logs of all user typing.
## Industry Reactions
- **Privacy Advocates:** Critical of the "friendly keylogging" label, noting that text-based tracking can capture passwords and sensitive data just as effectively as screenshots.
- **Market Response:** Professional users are intrigued by the automation potential, while enterprise IT managers remain wary of the data exfiltration risks.
## Future Outlook
- **Predictions:** Expect "Computer History" to eventually integrate with OpenAI’s "Operator" agents to perform tasks on behalf of the user based on observed patterns.
- **What to watch for:** Potential expansion to Windows and mobile, and the inevitable first report of sensitive data (passwords/PII) being sucked into a "Memory" summary.
## For Security Professionals
This feature significantly expands the **Attack Surface**:
1. **Prompt Injection:** Malicious instructions on a website can now be "read" by Computer History and executed by ChatGPT/Codex with context from other apps.
2. **Local Data Theft:** Since event logs are stored **unencrypted** for 48 hours, any local malware with user-level permissions can scrape a perfect record of the user's recent activity.
3. **Data Residency:** While OpenAI claims logs aren't used for training, the transition of "summarized memories" to the cloud creates new vectors for sensitive data leakage (PII/PHI).
4. **Governance:** Admins must proactively disable this feature in Business/Enterprise environments until a full risk assessment of "Accessibility API" abuse is completed.