After Microsoft invested and formed a partnership with the French language model startup MistralAI, the company stated that it continued to work hard to eliminate its image of hindering competition through deep cooperation with OpenAI (and financial investment). Today, the company launched a new framework called the "AI Access Principles" - an eleven-point plan that Microsoft says will "guide how we operate our AI data center infrastructure and other critical AI assets globally."
These points cover areas such as building and operating an app store where companies can pick and choose from different LLM and other AI products, and committing to keeping a company's proprietary data out of its training models. It also includes a commitment that customers will have the option to switch if they choose to switch cloud providers or in-cloud services. The report also details priorities for building cybersecurity around AI services, considerations for building data centers and other infrastructure in an environmentally friendly manner, and investments in education.
Microsoft President and Vice Chairman Brad Smith announced the framework today at Mobile World Congress in Barcelona. While this means Microsoft is open to dialogue and communication with stakeholders, Smith made the announcement during a keynote address and left no room for follow-up questions.
Microsoft's announcement comes as its $13 billion investment in OpenAI, a 49% stake in the startup that is leading the development of artificial intelligence services around the world, is coming under increasing regulatory scrutiny. In January, Europe's competition regulator said it was assessing whether the investment complied with antitrust rules.
Microsoft hopes to develop artificial intelligence as an important business area and enterprise services in the next few years, not just with operators participating in MWC, but with enterprises and organizations in more industries.
"If a customer trains a model on our infrastructure, we consider their data to be their data, and we do not access that data or use it to compete with companies that rely on our infrastructure," Smith said.
To be clear, these AI Access Principles are not binding rules for Microsoft -- nor are there any details on how any commitments will be verified or tracked -- but they have the intended effect. If there is any formal regulatory investigation, Microsoft may use these principles to demonstrate that it is actively working to ensure competition in the market.
"In fact, as of today, we have close to 1,600 models running in our data centers, 1,500 of which are open source models," Smith said onstage today. "That shows how we as a company ... focus on proprietary and open source models, companies large and small."
On the other hand, by such public articulation, the principles become a public statement that the public, Microsoft's competitors, and regulators can use as a benchmark if they think Microsoft is falling short.