Michael Blum left his post as UCSF's Associate Vice Chancellor for Informatics on what he calls a "calculated gamble." The bet: that healthcare organizations would pay to solve a problem that has vexed the industry for years—how to let artificial intelligence learn from patient data without actually exposing that data to anyone, including the people running the AI.
Three years after BeeKeeperAI raised $12.1 million in a Series A round led by Santé Ventures, that wager appears to be paying off. Though not spectacularly. Not yet, anyway.
The San Francisco company, which formally split from UCSF's Center for Digital Health Innovation in April 2022, has steadily assembled a portfolio of deployments that reads less like a consumer tech rocket ship and more like the deliberate, compliance-heavy grind of enterprise healthcare: oncology trial platforms in Atlanta, federal agency pilots, academic medical centers testing heart failure algorithms. All of it running on what BeeKeeperAI calls "sightless computing"—technology that encrypts health data even while it's being analyzed, rendering it invisible to the AI developers, the platform operators, and BeeKeeperAI itself.
It's an elegant solution to an intractable problem, assuming it can scale beyond the pilot-and-partnership phase that defines most healthcare AI ventures.
The Funding That Got It Started
When Santé Ventures led the June 2023 round—joined by the Icahn School of Medicine at Mount Sinai, AIX Ventures, Continuum Health Ventures, TA Group Holdings, and UCSF—BeeKeeperAI earmarked the capital for expanding its EscrowAI platform and broadening commercial operations. Dennis McWilliams, a partner at Santé, signed on to help guide the effort.
The company added at least one more financing event worth noting. In October 2024, Topcon Healthcare, a Tokyo-based ophthalmology technology giant, announced it would participate in a BeeKeeperAI funding round and that its CEO, Ali Tafreshi, would join the startup's board. The size and designation of the round were not disclosed. No further equity raises have surfaced publicly since.
For a startup operating in the capital-intensive, regulation-heavy world of healthcare infrastructure, the fundraising trail has been notably quiet. Whether that reflects BeeKeeperAI's ability to stretch its initial capital, a challenging fundraising environment for B2B health tech, or simply a preference for staying under the radar remains unclear.
What the Money Built
By the spring of this year, EscrowAI was processing AI models on protected health data across a handful of environments that, taken together, suggest a company feeling out multiple market pathways at once.
Massive Bio, an oncology trial-matching platform, deployed the technology in Atlanta to pre-screen patients for clinical trials within HIPAA-compliant environments, with a stated focus on underserved communities. The Icahn School of Medicine at Mount Sinai and Morehouse School of Medicine used EscrowAI in May 2025 to test AI models on multi-modal congestive heart failure data—a deployment that became the first certified under the Coalition for Health AI's assurance service provider process. An endorsement that carries weight in a field where trust and certification can unlock entire markets.
That same month, BeeKeeperAI partnered with cStructure to enable causal AI research, launching a "Covid Causal Diagram DREAM Challenge" that allowed researchers to compute on NIH COVID datasets inside EscrowAI's secure enclaves without ever touching the raw data. The move positioned the company as infrastructure for the kind of open science competitions that attract academic credibility.
The federal angle emerged in November 2024, when Government Acquisitions, Inc. integrated EscrowAI into its EdgeShield platform, creating a channel to U.S. government agencies. Intel's Federal CTO issued a statement. The door to defense and intelligence use cases—where privacy-preserving AI isn't just preferred but often mandated—cracked open.
Three months earlier, Topcon Healthcare had integrated EscrowAI with its Harmony platform, letting AI developers run models on real-world ophthalmology data. A pilot with AiScreenings followed, validating eye-disease detection algorithms on patient data confined within the secure enclave.
None of these deployments, individually, represents a market-defining moment. But cumulatively? They sketch a company methodically establishing credibility across academic research, clinical trials, federal procurement, and commercial partnerships. It's the slow build.
The Technology Underneath It All
EscrowAI runs on Azure Confidential Computing and Intel SGX secure enclaves, encrypting data and models at rest, in transit, and—here's the critical part—during processing. Most encryption schemes protect data when it's stored or moving between systems. The breakthrough, such as it is, lies in keeping it encrypted even when algorithms are actively analyzing it.
The platform uses attestation to verify the integrity of the computing environment before anything runs, and integrates Azure Confidential Ledger to create an audit trail that regulators can follow.
The technical foundation traces back to a 2020 collaboration among UCSF's Center for Digital Health Innovation, Fortanix, Intel SGX, and Microsoft Azure—work that established the confidential computing architecture before BeeKeeperAI existed as an independent entity. Blum, who holds an MD, took a leave from UCSF to lead the spin-out. He brought along Mary Beth Chalk as Chief Commercial Officer and Bob Rogers, PhD, as Chief Scientific Officer. Rogers previously served as Chief Data Scientist at Intel's Data Center Group, a pedigree that presumably helped navigate the technical partnership landscape.
The platform hit general availability in May 2023, a month before the Series A closed, and appeared in the Azure Marketplace with IP Co-Sell eligibility and support for Microsoft Azure Consumption Commitment deals—bureaucratic details that matter when selling into enterprises already committed to Microsoft's cloud ecosystem.
BeeKeeperAI holds multiple U.S. patents. The Coalition for Health AI's provider directory shows SOC 2 Type I certification with a Type II audit completed, though the company's Azure Marketplace page lists SOC 2 Type II certification. A minor discrepancy, perhaps, but the kind that raises questions about how current various public-facing information actually is.
Early Traction, Small Team
Before the flurry of deployments that followed the Series A, BeeKeeperAI had already worked with Novartis on two initiatives: a 2022 Biome Evidence Lab project applying sightless computing to pediatric rare disease algorithm development using UCSF data, and a 2023 "Beacon of Hope" collaboration focused on clinical trial diversity. Pharma partnerships that likely provided both revenue and validation.
The company's LinkedIn page indicates 11 to 50 employees, with 15 visible profiles as of mid-2026. That's a small team relative to the breadth of deployments scattered across academic medical centers, federal systems, and commercial partnerships. It suggests either a highly leveraged model—where the platform does most of the heavy lifting—or a company that's still very much in the land-grab phase, planting flags without the staff to defend them at scale.
Whether BeeKeeperAI can translate this collection of pilots and partnerships into the kind of recurring revenue that venture investors eventually demand remains the open question. Healthcare moves slowly. Especially when you're asking it to rethink how it handles its most sensitive asset.
For now, Blum's calculated gamble continues.
