OpenAI has unveiled new agentic capabilities for ChatGPT, enhancing its functionality by merging the user interaction of the Operator, deep research synthesis, and conversational intelligence. Users can now instruct ChatGPT to execute tasks and manage complex workflows, with built-in safeguards that request permission before significant actions. This allows for real-time user control, enabling interruption or task cancellation at any moment.
The updated ChatGPT agent integrates with various workflows using authenticated connectors, facilitating access to relevant information and allowing for the automation of recurring tasks. However, with these enhanced capabilities come new security risks, prompting OpenAI to bolster controls to safeguard sensitive data and ensure a broader user experience. The agent is trained to identify and combat prompt injections while implementing measures to correct model inaccuracies. With these updates, OpenAI emphasizes that this launch marks the beginning of ongoing improvements for ChatGPT’s agentic features.
Source link
