Assurance Resource Providers
Description
EscrowAI® is a CHAI-Certified Assurance Resource that enables privacy-preserving AI validation on primary clinical data—without data transfer or exposure. Built for both data stewards and model developers, EscrowAI ensures standards-aligned evaluation of model performance by protecting patient privacy, preserving intellectual property, maintaining data sovereignty, and enforcing transparent, auditable processes. Available globally via the Microsoft Azure Marketplace or via an on-premise deployment.
Timeline
TBA
EscrowAI is a CHAI-Certified Assurance Resource providing privacy-enhancing SaaS infrastructure that enables model developers to securely test and validate on primary clinical data curated to a specification from clinical environments.
It ensures that the AI development lifecycle—particularly data access, validation, and evaluation—is conducted with integrity, free from conflicts of interest, and in compliance with regulatory, ethical, and scientific norms.
For Data Stewards (e.g., AMCs and health systems): EscrowAI enables your organization to participate in the testing or validation of AI models while maintaining control of your clinical data. As a CHAI-Certified Assurance Resource, EscrowAI provides a secure, compliant infrastructure that allows external algorithm developers to send and test their models inside your HIPAA-compliant environment—ensuring your patients’ privacy is preserved, your data never leaves your control, and your organization remains aligned with ethical and regulatory standards.
EscrowAI protects against conflicts of interest, ensures your data is used as specified, and generates transparent, auditable reports—so you can contribute to advancing healthcare AI while maintaining complete trust and oversight. EscrowAI also enables data stewards to be recognized for their role in advancing responsible AI without compromising data governance and privacy.
For Algorithm Developers: EscrowAI offers a safe and trusted way to evaluate your AI models on high-quality, real-world health data from leading institutions—without exposing your model IP or needing to access raw patient data. As a CHAI-Certified Assurance Resource, EscrowAI automates and enforces strict privacy, conflict-of-interest, and data quality safeguards, ensuring your validation results are transparent, credible, and aligned with CHAI’s evaluation frameworks.
This enables faster deployment by providing credible, third-party validation of post-deployment model performance-without the traditional delays by data access approvals, custom integration, or regulatory concerns.
What Makes EscrowAI Different: EscrowAI enables secure, real-time validation on curated clinical data-delivering responsible, environment-specific performance insights while protecting privacy and intellectual property.
By aligning with CHAI’s consensus-driven standards, EscrowAI helps both data stewards and model developers build the transparency, accountability, and trust required to scale AI adoption responsibly in healthcare.
EscrowAI provides the following primary functions as a CHAI-Certified Assurance Resource:
For model developers:
Securely test and validate AI models on primary clinical data curated to their model’s specification
Receive attestation that the curated data meets defined data specification requirements
Protect algorithmic and data IP throughout the validation workflow
Establish an immutable, auditable record of each model execution, including model and data versions
For data stewards (including data controllers and processors):
Protect individual privacy without relying on de-identification
Retain full control of clinical datasets within their HIPAA-compliant environment (cloud or on-prem)
Eliminate secondary re-identification risk by preventing data movement or exposure
Support regulatory and ethical compliance through standardized, auditable workflows
Aptly named, EscrowAI enables collaborators to fully protect what they care about most in ways that result in optimizing model performance for the clinical environment.
The scope of the EscrowAI CHAI-Certified Assurance Resource includes:
In Scope
Independent AI Model Validation Facilitates unbiased, third-party testing of AI models without allowing data providers or model developers undue influence over outcomes.
Conflict-Free Data Collaboration Enables multi-party collaboration where sensitive data (e.g., patient-level clinical data) can be securely accessed without being moved, exposed, or manipulated.
Integrity Safeguards Ensures data integrity by enforcing that data shared is accurate, complete, and not selectively reported to skew results.
Transparent Disclosure Compliance Supports collection and disclosure of potential conflicts from participating entities (data providers, developers, researchers), as required under CHAI’s COI policy.
Firewall Enforcement Enforces operational and technical separation between parties involved in data supply and AI commercialization.
Regulatory Readiness Provides a compliant environment aligned with CHAI guidelines, helping AI solutions advance toward regulatory and clinical deployment.
Out of Scope
Monetization or Incentivized Outcomes EscrowAI is not intended for projects where financial incentives are tied to specific AI model outcomes or where biased data submission might confer competitive advantage.
Unvetted Partnerships or Undisclosed Relationships It cannot be used in contexts where partnerships between developers and data suppliers are not disclosed, as this would violate CHAI COI policies.
Control by a Single Party The resource must not be used in environments where the same entity controls both the data and the validation process without strict safeguards (e.g., firewalls, audits).
Unauthorized Data Use It does not permit the use of patient or proprietary data without explicit consent or compliance with legal standards, or compliance with applicable regulations (e.g., HIPAA, GDPR).
Conflict-Driven Operations Any willful or negligent manipulation of data or disregard for CHAI’s independence standards falls outside the scope and may result in loss of certification.
Artificial Intelligence Models
Machine Learning Models
Small- to Medium-Sized Large Language and Agentic Models (LLM)
Analytic Models
Data Queries
EscrowAI enables data stewards (both controllers and processors) to license clinical data sets for the purpose of testing and validating AI models. The EscrowAI confidential computing environment is compatible with any type of clinical or personal health data (including video, genomics, imaging, pathology, environmental, etc.), ensuring that algorithm developers may securely compute on real-world clinical data.
The model developer is responsible for specifying the data requirements detailed in the CHAI Assurance Resource Provider Appendix Figure 1 (pages 19-20), Figure 2 (page 20), Figure 3 (pages 21-22), and the Data Documentation Requirements (page 22) that are required for their model. The responsibility of attesting/validating that the requirements have been met lies with the data steward and is facilitated within EscrowAI.
Since the data remains with the data steward, they are responsible for creating and maintaining their organization’s data catalog. The EscrowAI platform facilitates secure computing access for model testing and validation, but does not aggregate or resell data. At no time does BeeKeeperAI take possession of the clinical data or the AI model.
EscrowAI is in active use by CHAI-aligned health systems and AI developers to validate models in privacy-preserving clinical data environments. Available globally via the Microsoft Azure Marketplace, EscrowAI enables scalable, standards-based assurance workflows that protect patient privacy and developer IP while accelerating responsible AI.
Icahn School of Medicine at Mount Sinai and Morehouse School of Medicine are currently utilizing the EscrowAI to establish a multi-modal cardiology data set that will initially support the testing and validation of chronic heart failure models.
EscrowAI uses a flexible, utilization-based pricing model that scales with the needs of the customer. Pricing is determined by the number of data computing locations and the project length (in months).
We offer tiered licensing to support a range of pricing structures—including on-demand, reserved, and enterprise.
EscrowAI supports enterprise-grade procurement through Azure Marketplace private offers, allowing bundled pricing for multi-party collaborations that can be applied to an organization’s Microsoft Azure Consumption Credit (MACC).
EscrowAI’s Project Opportunity Statement is a mutually approved document between the data steward and the model developer documenting the final scope of the project, including:
Model overview
Data specification
Output requirements
Acknowledgement of any conflicts of interest (COIs)
The Project Opportunity Statement is a key artifact of a project supported by EscrowAI.
The COIs for Data Stewards providing data (e.g., data controllers and/or processors) must disclose:
Financial interests, partnerships, or affiliations that could influence the quality or integrity of the data.
The data sources and methodologies are free from undisclosed conflicts.
Data suppliers do not have control over validation.
Entities involved in both data provision and model commercialization have implemented operational firewalls to separate their interests.
Model Developers must disclose:
Any relationships with data stewards that could bias outcomes. All potential conflicts will be: Declared in writing and submitted prior to mutually approving the Project Opportunity Statement.
Both the Data Stewards and the Model Developers share the following prohibited practices:
Withholding adverse data
Tying financial incentives to a favorable model outcome
Not disclosing partnerships between developers and data suppliers
Utilizing patient/proprietary data in a manner inconsistent with applicable laws and regulations
Should EscrowAI not enforce the data steward and model developer COIs, CHAI will have the right to:
First: Written warning with corrective steps to BeeKeeperAI
Second: Temporary de-certification (3 months) BeeKeeperAI
Third: Remove BeeKeeperAI from CHAI-Certified status (1 year before re-eligibility)
Icahn School of Medicine at Mount Sinai is an investor in BeeKeeperAI.
SOC 2 Type I Certified
SOC 2 Type II Audit Completed
EscrowAI is pursuing SOC2 Type II certification and anticipates full certification on or before August 1, 2025.
