OpenAI Launches "Dots": Agents Connected to 4,000 Apps

Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
OpenAI begins deploying autonomous agents called "dots" for paid ChatGPT offerings, featuring an Enterprise beta that can be activated by administrators and regional exclusions for European Pro subscribers. These agents operate on a cloud computer, access over 4,000 applications, and have operational safeguards in place. Specialized versions are being tested in businesses, and integration with Microsoft Agent 365 is underway.
Paid Access, Enterprise Beta, and Regional Restrictions
The dots are being rolled out in ChatGPT for Pro and Business Premium subscribers, with a gradual access that may take several days. Enterprise customers can access a beta version, provided that their workspace administrator activates it. For now, Pro users located in the European Economic Area, Switzerland, and the United Kingdom are not included, while OpenAI states that the Business Premium plan remains available in all regions where ChatGPT is accessible. The first dot is included in the subscription at no additional cost. OpenAI plans to later offer a paid option to add more agents or increase their workload capacity. Creating a dot can only be done on a computer, via the ChatGPT app or browser, before it can be used on mobile.
Autonomy Governed by Rules, Checks, and an Activity Log
Each agent operates in a sandboxed environment, hosted on a separate cloud machine from the user's, and utilizes saved passwords without the model having access to them. Default security measures are in place to prevent malicious instructions and detect risky behaviors; if a security incident occurs, the agent can halt or cease its activity. Authorization settings specify the situations where the agent acts autonomously and those requiring validation, with options ranging from total autonomy to user control. Sensitive operations related to accounts or data sharing are systematically verified, and certain actions, such as changing the password, remain the user's responsibility. An Activity interface allows users to monitor all tasks performed by the agent at any time and intervene to modify its course if necessary. OpenAI states that, by default, content from Business and Enterprise spaces is not used to train its models; for individual subscriptions, the user decides whether to participate in model improvement. OpenAI acknowledges that its agents can make mistakes and advises users to systematically check work that has significant implications.
Cloud Computer, 4,000 Apps, and Multichannel Interactions
The dots, based on GPT-6 Astra, continuously pursue the objectives set by the user and only solicit the user for validation or decisions, while learning their preferences over time. Each agent has its own cloud computer, equipped with a browser, and can write and test code, which is accessible to the user at any time. The dots connect to over 4,000 applications via the ChatGPT plugin ecosystem and enable interactions across multiple channels, including ChatGPT, Slack, and Teams, as well as through voice calls, maintaining context from one channel to another. OpenAI provides the example of a dot capable of identifying recurring requests in customer feedback, preparing and testing corrections and improvements, and then submitting complete pull requests accompanied by videos showing the changes. These agents were showcased during DevDay.
Enterprise Pilots and Integration with Microsoft
Specialized versions of the dots are offered to organizations, with their own identity, credentials, and access tailored to their mission. These agents adapt to new information and improve based on feedback from the teams overseeing them. OpenAI initially tested these agents internally in support functions such as procurement, billing, email marketing, customer support, and contract management. At this stage, only a few clients benefit from this access as part of pilot programs. OpenAI is also collaborating with Microsoft to enable the integration of these agents into Agent 365, allowing IT teams to manage them using existing tools.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.