SIAgentNews.com
← All news
DevelopingAgentsSecurity

Microsoft launches AI Transparency Initiative with 40+ industry partners

The initiative aims to create open standards for tracking how AI agents access data and take actions to ensure enterprise security.

· First reported by PR Newswire

40+number of organizations partnering with Microsoft

The short version

  • Microsoft is partnering with 40+ firms including Adobe, Mastercard, and Volvo Cars to create open AI transparency standards.
  • The initiative addresses the 'evidence gap' where organizations can see AI requests but not the internal logic between request and outcome.
  • Standards will apply to both closed and open-weight models to allow companies to swap providers without losing security oversight.
  • The project will be integrated into Microsoft AI, Microsoft Foundry, and Microsoft Copilot for customer use.

Why it matters

As companies give AI agents more autonomy to use tools and credentials, they need a way to audit those actions without relying on the AI's own word. This initiative helps create a universal 'trust architecture' so businesses can innovate without losing control over their data.

Closing the evidence gap

Organizations currently face a significant hurdle when deploying AI agents: they can often see what a user asks an AI to do, but they cannot easily see what happens inside the model to produce the final result. This lack of visibility makes it difficult for security teams to independently verify how a model behaves or why it made a specific decision.

This gap creates several risks for companies, including issues with security, compliance, and auditability. When AI agents are given the power to access internal data, tools, and credentials, they essentially act like privileged insiders. Without a way to see their internal reasoning, companies struggle to manage the risks posed by these increasingly capable systems.

Microsoft announced the AI Transparency Initiative at Microsoft Ignite to address these concerns directly. The goal is to move away from a 'trust the model' approach toward a system where every organization has greater visibility into the model itself. As Hayete Gallot, Executive Vice President of Microsoft Security, stated, "With this work, we will bring more transparency to all models."

A collaborative industry standard

Rather than creating a proprietary tool, Microsoft is working to establish open, industrywide standards for AI transparency and governance. The initiative involves a massive collaboration with more than 40 organizations, including major names like Accenture, Adobe, KPMG, Mastercard, and Volvo Cars. This broad participation ensures the standards will be useful across different sectors and technologies.

The initiative is structured into three distinct parts to move from theory to practice. First, Microsoft will expand its work with the National Institute of Standards and Technology's (NIST) SI Agent Standards Initiative. They will use research from this collaboration to develop the foundational rules for how AI systems should be governed and secured across various providers.

Second, the partnership with over 40 firms focuses on making AI systems accountable and safe by design. This means the standards are intended to work regardless of which specific model or provider a company chooses to use. By creating a common language for transparency, the initiative helps ensure that security controls remain effective even as the underlying technology evolves.

Mercury

How the framework functions

The initiative focuses on three core pillars: See, Verify, and Control. Under the 'See' pillar, organizations aim to know exactly what an AI was authorized to do, which model was used, and what specific data or tools it accessed. This provides a continuous monitor of risk signals against a company's established policies and safety thresholds.

The 'Verify' pillar is perhaps the most critical for security. It establishes that AI agents cannot be the sole source of truth regarding their own behavior. Instead, evidence of an agent's identity and delegated intent must be checked against independent records, such as identity and policy systems, rather than relying on the model's own self-reported signals.

Finally, the 'Control' pillar ensures that humans stay in charge of what an AI is permitted to do. By maintaining clear evidence of a workflow's purpose and owner, organizations can enforce policies before consequential actions occur. This evidence is designed to stay with the organization even if they decide to change models or providers in the future, maintaining a consistent audit trail.

Integration and adoption

Microsoft plans to bake these transparency standards directly into its core products. This includes Microsoft AI, Microsoft Foundry, and Microsoft Copilot. By doing so, they will allow customers to export the resulting evidence into their existing security and compliance tools, whether those tools are made by Microsoft or a third party.

This approach is designed to help developers move faster without sacrificing safety. Because the standards are intended to be provider-agnostic, developer teams can update the models powering their agents without breaking the security team's ability to govern them. It creates a modular way to handle AI security where the governance layer is separated from the specific model being used.

For large enterprises, this means they can finally begin to deploy AI at scale with a solid foundation of trust. As Harpreet Sidhu of Accenture noted, these types of open, industry-driven approaches help secure models and fit them with the necessary controls. It allows companies to move toward AI reinvention while keeping their boards, auditors, and regulators satisfied with the level of oversight provided.

What to watch

  • How quickly the 40+ partner organizations will adopt these standards into their own products.
  • Whether NIST will adopt these specific standards into their broader national guidelines.

Try it

Check out the official sign-up page at aka.ms/learnmoreAItransparency to see how you can get involved in shaping these standards.

Sources and links

First reported by PR Newswire. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news