OpenAI debuts AI personal assistant with file and browser control

OpenAI has introduced a new ChatGPT agent that functions as a personal assistant, capable of handling tasks such as finding restaurant reservations, shopping, and even drafting job candidate lists. The AI agent, now available globally except in the EU, combines AI research tools with the ability to control web browsers, computer files, and software like spreadsheets and slide decks.

Key Features and Risks

• Autonomous Task Handling: The agent can autonomously complete assignments like drafting travel itineraries or conducting work research by toggling between systems.
• User Control: OpenAI emphasizes that users remain in control, with the agent requesting permission before taking any destructive or irreversible actions.
• Increased Risks: OpenAI acknowledges that this model carries more risks than previous versions, including potential misuse for creating biological hazards. Safeguards have been implemented to mitigate these risks.

Commercial Implications

The launch has sparked speculation about OpenAI's revenue potential, particularly if the agent guides users to retail checkouts. CEO Sam Altman has previously suggested a 2% fee on sales generated through its "deep research" software. However, OpenAI confirms the agent does not include sponsored product recommendations.

Demonstration and Safety Measures

A recent demo showed the agent:

1. Checking a user’s Google Calendar for free slots.
2. Searching for highly-rated restaurants (Italian, sushi, or Korean) with at least 4.3 stars.
3. Providing options within 10-15 minutes.

Safety checks include training the agent to reject suspicious requests, such as demands for bank transfers. The system will initially be available to paid subscribers of the "pro," "plus," and "teams" versions.

Industry Context

OpenAI's move follows similar launches by Google and Anthropic, reflecting growing interest in AI agents that can autonomously handle computer-based tasks. Analysts like Niamh Burns of Enders Analysis note the potential for ads or sponsored placements as AI companies face pressure to monetize.

For more details, check out these related articles:

• UK’s £225m AI supercomputer, Isambard-AI, launches in Bristol
• AI firms ‘unprepared’ for dangers of building human-level systems, report warns

"The hope is that agents are able to bring some real utility to users – to actually do things for them rather than just outputting polished text and sounding impressive," said Burns.