The national science agency says the next hurdle is the TGA, not the technology.
Artificial intelligence in healthcare is rapidly moving beyond research projects and proof-of-concept demonstrations into regulated clinical products, with the CSIRO warning developers that success will increasingly depend on meeting medical device standards rather than simply building better algorithms.
In its new AI Trends for Healthcare 2026 report, the CSIRO said many AI applications intended for clinical use fell within Australia’s Software as a Medical Device (SaMD) framework, making regulatory strategy, quality management systems, and Therapeutic Goods Administration approval critical components of development rather than afterthoughts.
“This 2026 report features two current issues for safe and responsible AI. The first is quality management, an often underestimated and under accounted aspect of technology implementation,” the report said.
“Many AI-based tools qualify as Software as a Medical Device (SaMD) under government regulation.”
This article originally ran on TMR’s sister site, Health Services Daily. TMR readers can sign up for a discounted subscription.
The report noted that CSIRO’s Australian e-Health Research Centre recently completed the Medical Device Single Audit Program audit and “was recommended for certification, enabling us to progress our solutions through regulatory submissions and accelerate market access for software-based medical devices in medical imaging analysis and clinical decision support with the TGA, FDA, and Health Canada”.
The CSIRO argued that quality management should be embedded from the beginning of product development rather than bolted on immediately before regulatory approval.
“A full section on the importance of quality management systems from the very early stages of software as a medical device development (SaMD), as a means of ensuring safe and responsible AI, is outlined in this report,” it said.
The report devoted an entire chapter to what it described as moving AI “from code to care”, arguing that regulatory compliance was fundamental to translating promising research into products clinicians can safely use.
“For innovations to move from research into clinical practice, software as a medical device (SaMD) must have approval from the Therapeutic Goods Administration (TGA),” the report said.
It added that “a robust quality management system (QMS) is a significant component in achieving TGA approval and ensuring research innovations can be translated into real-world healthcare benefits for patients”.
Rather than treating regulation as the final stage of development, the CSIRO said regulatory planning should begin alongside research.
“While most companies only develop their QMS before applying for a conformity assessment, the AEHRC treats QMS as a continuous process, one that should be considered even before R&D begins,” the report said.
“When an idea is first formulated, a quality manager will join the team to develop a regulatory strategy – how to ensure that the product’s healthcare benefits can be realised as soon as possible.”
Related
The report also identified traceability as a key element of trustworthy AI development, describing it as “a core element of quality management” that links user needs through to software requirements, implementation and testing.
The CSIRO said embedding these governance processes throughout development was essential if healthcare AI was to gain clinician confidence and successfully navigate regulation.
“Through our commitment to quality management, we operationalise responsible AI principles across the entire SaMD lifecycle,” the report said.
“This ensures that our AI-enabled healthcare technologies are not only innovative, but safe, trustworthy, reliable, and ready for regulatory approval and clinical impact.”
The report revealed how AI was now embedded in real-world clinical settings for more connected, safer and more efficient healthcare.
“For many years, AI has largely been ‘under the hood’– a powerful but often invisible technology understood mostly by technical experts,” said Dr David Hansen, CEO and research director of the AEHRC.
“The rapid rise of generative AI has changed that. These tools have brought AI into the spotlight and accelerated its integration into healthcare – while also sharpening the focus on safety, quality and responsible use,” he said.
The report demonstrated how AI was already delivering measurable benefits across healthcare, from clinical decision support and medical imaging analysis to disease management and personalised care.
In one case study, AI was being used to generate synthetic CT scans from MRI images, helping clinicians plan more accurate radiotherapy treatment and reducing patients’ exposure to radiation.
As healthcare organisations, researchers, industry and governments adopted AI, the report emphasised realising AI’s full potential depended on strengthening the systems that underpin it.
“As healthcare systems increasingly rely on AI-powered tools, the need for robust evidence, quality assurance and community co-designed standards has never been greater,” said Dr Hansen.
The report also examined emerging technologies such as multimodal AI and AI-assisted software development, offering insight into the innovations shaping the next generation of healthcare.
“We found new AI technologies need to be developed hand-in-hand with clinicians and industry. The research in this report signifies rich collaboration and therefore equally rich real-world utility,” Dr Hansen said.
The report identified key challenges that must be addressed to ensure AI technologies are safe and effective at scale including regulation, quality management, data governance and digital health interoperability.
It also outlined the critical role of national digital health standards, including work underway through Sparked, Australia’s FHIR accelerator, to ensure new AI technologies can integrate effectively across the healthcare system and support connected, patient-centred care.
With more than two decades of experience developing healthcare AI, Dr Hansen said the sector was entering a new critical phase.
“We are in a pivotal new chapter where responsible innovation, rigorous evidence and collaboration will determine how successfully AI delivers on its promise for patients, clinicians and communities.” Dr Hansen said.
