Bunkerhill Health

The product is AI infrastructure: a platform whose entire function is building, deploying, and governing AI agents, plus the company's own FDA cleared imaging algorithms that run inside it. Remove the AI and nothing remains.

Autonomy is claimed and oversight is asserted without a mechanism. Bunkerhill sells outcomes ('Name the outcome. Bunkerhill delivers.'), and its UTMB case study reports 20+ live agents with '100% of agentic workflows fully automated', including agents that flagged patients with incidental coronary calcium and initiated cardiology referrals.

UTMB's AI officer describes AI as a partner, not a replacement for human judgment, but no review step, escalation threshold or limit on what an agent may do is described. An agent authoring platform puts oversight design on the customer, and nothing published says what the platform enforces or which agent actions require clinician sign off.

Two levels of disclosure inside one product. Bunkerhill's cleared imaging algorithms (coronary and aortic valve calcium, mitral annular calcification, bone mineral density, abdominal aortic diameter) have 510(k) summaries describing intended use, study design and performance. The agents and chat that make up Carebricks are described by what they do, reading imaging, reports and the full record and taking actions, with no model, version or evaluation published.

The architecture is described and no model provider is named. Bunkerhill's agents read imaging archives, report archives and the full record and act across a health system; its own imaging algorithms are cleared devices, but the language model behind the agents and chat, the host and any subprocessor are not named, and no subprocessor list appears outside the script rendered trust center.

Named deployments with outcome figures, short of published cohort results. Bunkerhill's case studies report UTMB Health running 20+ agents across 7+ clinical domains, 80 percent faster lung nodule triage there, and 11.7 percent of routine chest CT patients found with unaddressed coronary calcium; Endeavor Health identifying 5.5 percent of routine chest CT patients for cardiac follow up; WVU improving early coronary disease detection by 7 percent; and HCA West Florida detecting coronary calcium in 58 percent of chest CT scans. The cleared algorithms have 510(k) performance summaries; the agent outcomes have no stated method or baseline.

General assurances that do not answer the questions AI raises. Bunkerhill states HIPAA compliance, SOC 2 Type II and ISO 27001, and links a trust center that may hold its policies. Its agents read imaging archives, report archives and the full longitudinal record across a health system. Nothing on the product pages states retention periods, deidentification, or whether customer data trains the agents or algorithms. Safety engineering is not published.

Compliance is claimed without the document behind it. The home page states Bunkerhill is fully compliant with HIPAA privacy and security standards, alongside SOC 2 Type II and ISO 27001, and links a trust center. No business associate statement or agreement appears on the company's own pages, though the trust center may hold one.

Recognized certifications are named with their type, and the artifacts sit in a trust center. The home page states SOC 2 Type II, ISO/IEC 27001:2022 certification and HIPAA compliance and links a trust center on a compliance automation platform. The trust center may hold the ISO certificate scope and the SOC 2 report period; nothing outside it states them.

The product the record leads with has no device position, while the company's own imaging algorithms are cleared. Carebricks, Bunkerhill's agent platform, runs agents for referral triage, lung nodule follow up, heart failure follow up and prior authorization with no FDA clearance and without the cleared label; a UTMB case study says it automatically initiated cardiology referrals from calcium findings.

Bunkerhill's algorithms hold 510(k) clearances in the public database, including Contrast CAC (K260166) and AVC (K243229), and the home page attaches the cleared label only to its bone mineral density measurement, correctly. CMS set a national billing code for algorithmic CAC and AVC analysis on chest CT from April 2026.

Governance is asserted rather than documented. Bunkerhill has described governance, monitoring and safety capabilities built into every deployment; its current home page and case studies do not repeat that, and the trust center may hold an AI policy. No framework, bias evaluation or subgroup results are published. For a platform that fully automates workflows across clinical domains, nothing published describes a review of each agent before it goes live or who signs it off.

Mechanisms exist for part of the product. The cleared imaging algorithms carry 510(k) summaries with performance and study design, plus the complaint and reporting duties of a cleared device, which gives a buyer something to hold those measurements to. The agents that act on them, initiating referrals, drafting documentation and working prior authorizations, carry no published accuracy figure, audit trail description, warranty or remediation commitment. One product carries two evidence regimes, and nothing published says which agent actions are logged or what Bunkerhill commits to when an agent acts wrongly.

Integration is central and no system is named. Bunkerhill's agents read clinical report and imaging archives and the EHR, and act by generating documentation, notifying care teams, initiating referrals and patient outreach, as the UTMB case study describes. No EHR, standard, listing or write back specification is named.

A hosted service with nothing committed about location. Bunkerhill runs Carebricks as a platform connected to a health system's imaging and report archives and EHR; no hosting provider, region, tenancy model or residency commitment appears on the home page or case studies, though the trust center may hold them.

No public pricing. Sold through enterprise agreements with health systems; no rate card published. The CMS billing code for CAC and AVC analysis creates a reimbursement path for certain algorithm use, which is a distinct economic question from platform licensing cost.

Coverage is broad by design, spanning cardiology, nephrology, pulmonary nodule follow up, oncology prior authorization, and registry operations across clinical, operational, and administrative functions. Clearly described, though breadth is a property of the platform model rather than a validated scope.