Microsoft Research: Vega — ZK proofs for digital identity, 92ms generation and 70% faster repeated proofs

Microsoft Research presented Vega on 21 May 2026 — a zero-knowledge proof system designed for digital identities in the age of AI agents. Vega solves a concrete, long-standing problem: how to prove a fact from a government document (driving licence, passport, ID card) without revealing the document itself or all the data it contains.

What are the concrete performance figures of the Vega system?

Vega achieves the following metrics on standard consumer devices:

• 92 ms for proof generation
• 108 KB proof size
• 23 ms for proof verification
• Up to 70% faster repeated proofs through the fold-and-reuse technique

Implemented in Rust, it is compatible with the EU Digital Identity Wallet (EUDIW) standard and mobile driving licences (mDL) that use a 2KB base document size. These performance figures are significant because previous ZK proof systems for identity had latencies on the order of seconds — too slow for interactive web or mobile use cases.

How does Vega technically achieve these performance figures?

Microsoft Research highlights two key innovations:

Fold-and-reuse proving — in traditional ZK systems, each proof is generated from scratch. Vega caches intermediates from the first proof and uses them to accelerate repeated proofs of the same credential. Concretely, if a user proves their age multiple times within a month, the second and third proofs will be 70% faster than the first.

Lookup-centric circuit design — classical ZK proof systems for documents parse the entire CBOR (or JSON) document inside the ZK circuit, which is computationally very expensive. Vega uses a lookup table approach where the circuit accesses only the specific fields needed for the particular proof, avoiding the entire parsing overhead.

Why is Vega particularly important for AI agents?

Microsoft explicitly connects Vega to AI agent use cases. The reason is concrete: if an AI agent needs to prove a user’s identity to a third party — for example, confirming age for a purchase, verifying qualifications for access to a particular service, or confirming residency status — current approaches require the agent to have access to a complete copy of the government document.

This is dangerous for multiple reasons. First, the agent can be compromised (prompt injection attack) and the document exfiltrated. Second, agents typically live in cloud environments where storing sensitive data introduces compliance problems (GDPR, HIPAA). Third, users are reluctant to give an AI agent access to original ID documents.

Vega solves the problem differently — the user generates a ZK proof on their device (where the document lives), then gives the agent only the proof. The agent can then submit the proof to a third party that verifies it in 23ms, without ever having seen the document.

How does Vega fit into the broader ZK ecosystem?

Vega arrives in the context of growing activity around digital identity standards. The EU Digital Identity Wallet (EUDIW) is in the pilot deployment phase in several member states. The ISO 18013-5 mDL standard is increasingly being adopted in the United States and other countries. Apple, Google, and Microsoft are all increasingly integrating digital ID into their mobile platforms.

Existing ZK tools (zk-SNARKs, zk-STARKs, Halo2) are powerful but generic. Vega is specialised for the identity use case with concrete performance targets. This may mean Microsoft is aiming to standardise its own variant through the EUDIW Reference Implementation or other industry forums.

For enterprise users, Vega opens up the possibility of building agentic workflows that work with regulated data (healthcare, finance, government) without compromising user privacy. It is worth watching whether Microsoft decides to open-source the Vega implementation or offer it through an Azure managed service.