Harrison.ai, the Australian developer of radiology and pathology AI, is now supporting Frontier Radiology, a US teleradiology group that calls itself “AI-native” and is actively hiring radiologists. The move, reported by The Imaging Wire, stretches the company’s business model from selling algorithms to underpinning a clinical service. It also drew a critical report from Australia’s ABC News questioning layoffs at home, the fate of public money and a potential conflict of interest.
What Frontier Radiology Is and What Harrison Provides
According to The Imaging Wire, Frontier is owned and operated by radiologist Joshua Ewell, DO, in line with US rules that require physician ownership of entities delivering clinical services. Ewell’s LinkedIn profile describes Frontier as an “AI-native radiology group” built entirely around foundation models.
Harrison’s role is non-clinical. It supplies its suite of algorithms, which the company says already helps Frontier’s teleradiologists reach some of the fastest report turnaround times in the industry. Frontier’s radiologists, Harrison says, remain free to use any AI they choose, including competitors’ products. Job ads cited by ABC described Frontier as Harrison’s “AI-enabled teleradiology service”; the group’s website, per the broadcaster, presents it as an independent medical practice that is “affiliated” and “partnered” with Harrison, which provides administrative, operational and technology support.
The ABC Report: Layoffs, Public Money and Conflict of Interest
ABC’s story, published September 8, started from a contrast. In 2025 the Australian government’s National Reconstruction Fund invested A$32 million (about US$23 million) in Harrison, in a round that valued the company at roughly A$400 million. At the time, fund chair Martijn Wilder said the investment would ensure the company kept its operations based in Australia. In April 2026, Australian staff were told of redundancies; ABC could not establish how many people were let go, but found at least two former employees who announced their departure on LinkedIn in May.
The broadcaster also described an internal restructuring meant to turn every employee into a “player-coach” of AI agents, quoting Matt Geleta, the company’s director of AI operations, as saying some people chose to leave in search of more traditional workplaces. The sharpest point was conflict of interest. Michelle Lazarus of Monash University warned of automation bias: “if you’re just following the AI recommendation and not challenging it, you can put health at risk.” Wendy Rogers, a clinical ethics professor at Macquarie University, said doctors employed by the tool’s provider could struggle to push back on its output. ABC also noted that Frontier is registered in Delaware with undisclosed ownership, and that a pay calculator on its website offers a 25% individual bonus for AI-assisted productivity gains, assuming a 30% productivity increase. Harrison, ABC said, did not respond to its requests for comment.
Dimitry Tran’s Response
Speaking to The Imaging Wire, co-founder and CEO Dimitry Tran, who relocated to California to lead the US push, disputed that narrative. In his view, ABC conflated unrelated developments. The layoffs earlier this year stemmed from the company’s shift out of an aggressive R&D phase into commercialization and had nothing to do with US entry. The Australian government’s money, he added, bought a single-digit equity stake that the fund still holds and that should grow more valuable as Harrison succeeds.
Tran also argued that tight cooperation between AI developers and imaging providers is no longer unusual. Two of the largest US providers, Radiology Partners and RadNet, have built their own AI divisions serving both in-house radiologists and outside customers, the mirror image of Harrison’s move from software toward clinical operations. The company, he said, will keep its focus on developing and commercializing algorithms, above all the Harrison.Rad foundation model, while keeping the “vast majority” of its staff in Australia.
Harrison.Rad 1.5 and the AI-Provider Convergence
The technical core of the strategy is Harrison.Rad 1.5, a foundation model released in June that drafts reports from images, priors and clinical context for radiologist review. Per the company’s technical report, it covers chest, musculoskeletal, abdominal, spine and pelvic radiographs plus mammography, and its vision encoder was trained on roughly 6 million image-report pairs. Harrison claims it is the only model to have passed the FRCR 2B Short Case exam used to certify UK radiologists. That is a company claim, not yet independently validated in the peer-reviewed literature.
A teleradiology group built around such a model is, in effect, a scale laboratory: every reviewed report yields usage data and feedback on the draft. That is what makes the relationship valuable to Harrison, and precisely what worries critics, because the vendor’s commercial incentive now sits next to the physician’s clinical judgment. Governance of imaging AI is already on the agenda of the big US groups, as Radiology Partners’ request for clearer FDA rules showed, and Rad Partners’ US$715 million purchase of Everlight signals that teleradiology has become a strategic asset.
What It Means for Radiology Worldwide
The backdrop is a workforce that cannot keep pace with imaging volume. In the US, the Veterans Affairs system lost roughly half of its teleradiology capacity, as we reported this month, and health systems from the UK to Brazil lean on remote reading for nights, weekends and rural coverage. The plumbing is in place: cloud teleradiology already unifies distributed worklists, so wiring draft-generating AI into that flow is the obvious next step. Physician-ownership rules like those in the US, and technical-responsibility rules such as Brazil’s CFM Resolution 2,107/2014, will shape how far vendors can go in operating clinical services rather than merely supplying them.
There is history here too. Harrison grew up alongside I-MED, Australia’s largest imaging network and an investor in the company, sold to Jardine this year, and trained models on millions of de-identified I-MED scans without notifying patients. Per ABC, Australia’s privacy regulator closed the matter without sanction.
Outlook and Limitations
The Imaging Wire’s read is that, spin aside, Harrison’s teleradiology venture is an intriguing twist on the growing integration between AI and imaging providers, perhaps the most direct route to relief for overstretched radiologists. Open questions remain. Radiologist independence from a commercial partner’s tool will need verifiable safeguards, not just the formal freedom to use rivals. And chest X-ray algorithms, as independent studies show, perform unevenly across products and populations. A company that both runs the clinical service and sells the AI will have to prove, with public data, that drafts speed up reporting without nudging physicians to sign what they have not fully read.
Source: The Imaging Wire · ABC News
