On August 29, 2025, OpenAI announced the acquisition of an American startup specializing in artificial intelligence applied to intelligent agents, for an estimated $1.1 billion. This is one of the largest acquisitions the company has ever made, and the third-largest acquisition in the AI sector since the beginning of the year1.
This announcement comes as OpenAI is stepping up the integration of its models into practical applications, such as ChatGPT, custom GPTs, voice assistants, and its productivity tools embedded in Microsoft Copilot. More than just an acquisition, the deal symbolizes a shift in scale for OpenAI: it is no longer just about developing cutting-edge models like GPT-4 or GPT-5, but about mastering the entire application chain, from infrastructure to final deployment.
As the use of artificial intelligence becomes increasingly widespread, this move raises several questions: What is OpenAI really aiming to achieve with this acquisition? What concrete impacts will it have on its products, particularly ChatGPT? And what ethical implications can we anticipate as AI becomes ever more powerful, autonomous, and integrated?
A strategic acquisition worth $1.1 billion
According to documents filed with the SEC2 and sources familiar with the matter, OpenAI has finalized the acquisition of Multi (formerly MultiOn), a California-based startup founded in 2022 that specializes in developing AI agents capable of interacting with web applications like a human user. Multi had raised more than $20 million from investors such as General Catalyst and Lightspeed Venture Partners before being acquired.
This $1.1 billion acquisition underscores the strategic value placed on Multi’s expertise in automated task execution on graphical user interfaces—a key area for evolving ChatGPT from a conversational model into a true executable intelligent assistant capable of booking a ticket, responding to emails, or navigating a website on the user’s behalf3.
OpenAI is no longer hiding its ambition: to transform ChatGPT into a multi-purpose AI agent, somewhere between a software co-pilot and a universal personal assistant. This acquisition would accelerate the implementation of this vision by leveraging technology capable of interpreting web interfaces, detecting input fields, clicking, filling in forms, and executing actions—all in a generic manner, without the need for custom development.
ChatGPT and Beyond: What This Acquisition Will Change
With over 100 million monthly active users on ChatGPT, according to OpenAI, the challenge now is to translate this success into concrete, sustained, and monetizable use cases. By adding the ability to take action on the web or within third-party applications, the AI agent is taking a new step forward: it no longer merely responds—it takes action.
OpenAI could thus:
- integrate an executable assistant feature into ChatGPT (fill out forms, send emails, interact with interfaces),
- enhance the alignment between memory, user intent, and task execution,
- provide advanced services to businesses through the GPT API with autonomous actions.
This move is part of a trend toward the platformization of AI, in which ChatGPT would no longer be a standalone product but rather an intelligent hub capable of interacting with other software, adapting to the user’s context, and learning from past interactions. All of this would be powered by GPT-4o or GPT-5 models and Microsoft Azure infrastructure.
Ethics and Governance: Increasingly Powerful AI—For Whom and How?
This rise of autonomous AI raises major ethical questions. Who controls what the agent does? At what point can it act on its own? How can we ensure that it does not carry out undesirable actions?
With AI capable of clicking, navigating, and interacting on our behalf, questions arise regarding digital consent and the interpretation of intentions. If the agent makes a mistake, who is responsible: the user, the interface designer, or the model provider?
These concerns echo those raised by European regulators in the context of the Artificial Intelligence Act (AI Act), particularly regarding general-purpose systems combined with critical functions4.
OpenAI has announced that the agents resulting from this acquisition will operate using explicit control mechanisms, restricted execution environments (sandboxes), action logs, and a strengthened user authorization system. However, the governance of these systems, their auditability, and their impact on human autonomy will need to be closely monitored.
Industrial Strategy: Is OpenAI Aiming for Full Integration?
The acquisition of Multi for over a billion dollars is part of a strategy of full vertical integration. OpenAI is no longer content with simply developing language models; it now seeks to control the entire value chain of generative AI: fundamental research, execution infrastructure, application interfaces, and customer relations.
The company already works closely with Microsoft to host its models on Azure, but is also developing its own tools (ChatGPT, GPTs, personalized assistants, contextual memory tools). With this acquisition, it is adding a strategic functional component: the ability for its models to take action in heterogeneous software environments through a dynamic understanding of interfaces.
In short, OpenAI is no longer content with simply developing AI; it is positioning itself to shape its applications through native apps, intelligent APIs, and, in the future, interconnected executable agents.
This move is not an isolated one. It comes in response to the rise of competitors such as Google (with Gemini and its agents integrated into Workspace), Meta (with its multimodal models), and even Anthropic and Amazon. But OpenAI is making a major move by acquiring a company entirely focused on agent-based execution, with technology that is difficult to scale.
Toward Applied, Proactive… and Invisible AI?
With this acquisition, OpenAI is no longer merely seeking to improve the quality of its responses, but to transform the very nature of human-machine interaction. By combining language models, contextual memory, perception (through vision and audio), and autonomous execution, AI becomes a proactive assistant capable of anticipating and taking action.
This development opens up exciting possibilities, but also raises concerns: How can we maintain control? What role should error, human oversight, and explainability play? And to what extent will we allow AI to “do things for us” without losing our digital autonomy?
OpenAI’s acquisition of Multi, in a deal worth over $1 billion, marks a new milestone: that of applied, integrated, and actionable artificial intelligence, which is redefining the boundaries between tools, assistants… and digital agents.
Learn more
On this blog, we’ve already explored the evolution of intelligent agents and OpenAI’s strategy:
ChatGPT Agent: OpenAI Introduces an AI Capable of Planning, Executing… and Learning
OpenAI introduces “connectors”: toward a ChatGPT integrated into the work environment
References
1. OpenAI. (2025). OpenAI acquires Multi to accelerate agentic AI.
https://www.mistral.ai/
2. U.S. Securities and Exchange Commission. (2025). Form 8-K: OpenAI Inc. acquisition disclosure.
https://www.sec.gov/
3. CB Insights. (2025). Global AI Deals Report – Q3 2025.
https://www.cbinsights.com/research/report/ai-deals-q3-2025
4. European Commission. (2024). AI Act: Guidelines for high-risk and general-purpose AI systems.
https://digital-strategy.ec.europa.eu/
