Brief IA
›

Microsoft Brings Together Over 40 Organizations for Verifiable AI

Business & Startups·Tom Levy·

Microsoft Brings Together Over 40 Organizations for Verifiable AI

Microsoft Brings Together Over 40 Organizations for Verifiable AI
Key Takeaways
1Microsoft launches a transparency initiative with over 40 organizations.
2The group plans to integrate the work into Microsoft AI, Microsoft Foundry, and Microsoft Copilot.
3The system is designed for various models and providers.
Why it matters : Microsoft aims to enable companies to independently verify the activities of agents using their own security and compliance tools.

Microsoft launches the AI Transparency Initiative with over 40 organizations, aimed at establishing a common approach to transparency, security, and governance of models and agents. Announced at Microsoft Ignite by Satya Nadella, CEO of Microsoft, it is based on a clear requirement: increased autonomy must come with greater transparency.

Companies Remain Responsible for Agents

Access to data, tools, and identifiers gives agents a privileged profile. Their operation at the speed and scale of machines makes independent oversight essential, without transferring responsibility outside the organization. If a significant event occurs, its leaders must explain themselves to customers, employees, the board of directors, auditors, and regulators.

For Microsoft, an agent should not be the sole witness to its own behavior, and the company must retain control over the actions it can undertake. The group believes that evidence regarding the purpose, responsible party, data, and delegated powers of each process helps enforce internal rules before any significant action takes place.

Microsoft Aims to Make Actions Verifiable

Companies are already linking models to their data, tools, identifiers, and infrastructures, allowing agents to act without seeking human validation at every step. Control mechanisms track a large portion of this activity, but not the internal operations of the models. Organizations often cannot reconstruct the pathway between the request and the response, complicating independent verification. The lack of visibility into the inner workings of the models also limits understanding of the context of agents' actions.

La lettre du soirGratuit · 19 h

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.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Hayete Gallot, Executive Vice President of Microsoft Security, believes that better visibility into the model itself is necessary for effective controls. Microsoft wants each organization to be able to trace the permissions granted, the model used, the execution environment, the data and tools consulted, as well as the actions taken. This verification should be possible independently, using the company's own security and compliance systems.

Microsoft Plans Common Open Standards

Accenture, Adobe, KPMG, Mastercard, and Volvo Cars are among the participants. The program consists of three components, including Microsoft's expanded participation in the SI Agent Standards Initiative to develop open industry standards. Cooperation with partner organizations is also expected to produce standards for responsible, independently verifiable, and securely designed systems.

Microsoft promises better visibility into the interactions between AI models and agents and corporate data in its products and services. The group plans to incorporate the work into Microsoft AI, Microsoft Foundry, and Microsoft Copilot; customers should be able to leverage the evidence obtained with their security and compliance tools.

The framework is also designed for multiple models and providers. The stated goal is to allow developers to replace or update models without compromising the ability of IT and security teams to govern them. Interested organizations can access additional information and track the progress of the program.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.