Overview

• Operator’s standalone service is being phased out as its browsing, terminal and API tools integrate directly into ChatGPT’s main interface.
• ChatGPT can now proactively perform multi-step tasks on a virtual computer: web navigation, code execution, API calls.
• Advanced agentic functions are limited to Pro, Plus and Team subscribers, with Pro users capped at 400 agent prompts per month and smaller allocations coming for other tiers.
• OpenAI has implemented threat modeling, prompt-injection detection, explicit user confirmations, active supervision to address increased privacy and security risks.
• Early demonstrations reveal the agent can take excessive time for simple requests and produce errors, reinforcing the need for user oversight.