ChatGPT Agent: Your New AI Co-Worker with Its Own Computer!

OpenAI launches ChatGPT agent, a new system merging Operator's remote browser, deep research capabilities, and ChatGPT's conversational skills. It rolls out to Pro users by day's end, Plus/Team users in days, and Enterprise/Edu users in weeks, offering tools like browsers, terminal, and APIs for diverse tasks.

Key Capabilities

• Tool Integration: Uses visual/text browsers, terminal (for code execution, data analysis, slides/spreadsheets generation), and direct APIs. For example, it can run code to analyze market data and auto-generate a spreadsheet with insights.
• Task Autonomy: Chooses optimal paths dynamically, maintaining task context across steps. It excels in academic and real-world tasks like data modeling and investment banking analysis.

Safety Measures

• Risk Mitigations: Addresses new risks, especially prompt injection (a common threat to agentic systems) with extensive safeguards.
• High-Capability Classification: Treated as High Capability in Biological and Chemical domains under OpenAI's Preparedness Framework (a precautionary step, even without definitive evidence of enabling severe harm). This activates strict safeguards, as outlined in its system card.
• Preparatory Efforts: Builds on earlier work in AI biology safety, including training models to handle harmful requests, deploying detection systems, and collaborating with experts (e.g., hosting a July biodefense summit).