--- title: "Risk-based software rules, with AI evidence rules layered…" description: "Australia's TGA reformed its software-as-a-medical-device rules in 2021 and has since been bolting AI-specific evidence expectations onto that framework…" lang: en json-ld: | [ { "@context": "https://schema.org", "@graph": [ { "@type": "WebSite", "@id": "https://ai-samd.com/#website", "url": "https://ai-samd.com/", "name": "AI SaMD Playbook", "description": "Working reference for regulating AI/ML Software as a Medical Device: FDA, EU AI Act, MHRA, Health Canada and NMPA guidance, model risks, and cybersecurity for RA/QA teams.", "inLanguage": "en" }, { "@type": "Organization", "name": "Blue Goat Cyber", "url": "https://bluegoatcyber.com" } ] }, [ { "@context": "https://schema.org", "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://ai-samd.lovable.app/" }, { "@type": "ListItem", "position": 2, "name": "Crosswalk", "item": "https://ai-samd.lovable.app/crosswalk" }, { "@type": "ListItem", "position": 3, "name": "Risk-based software rules, with AI evidence rules layered on.", "item": "https://ai-samd.lovable.app/regulators/tga" } ] }, { "@context": "https://schema.org", "@type": "Article", "headline": "Risk-based software rules, with AI evidence rules layered on.", "description": "Australia's TGA reformed its software-as-a-medical-device rules in 2021 and has since been bolting AI-specific evidence expectations onto that framework rather than writing a stand-alone AI regulation. The 2024 evidence guidance for AI-enabled software, plus the Feb 2026 refresh of the AI and medical-device software page, are now the operational reference points for any AI SaMD entering the ARTG.", "about": "Australia · Therapeutic Goods Administration", "mainEntityOfPage": "https://ai-samd.lovable.app/regulators/tga", "publisher": { "@type": "Organization", "name": "Blue Goat Cyber", "url": "https://bluegoatcyber.com" } } ] ] --- [AI SaMD Playbook ](/) Sections[Crosswalk](/#crosswalk)Tools Search Ctrl K[Talk to us →](https://go.bluegoatcyber.com/meetings/blue-goat-cyber/discovery-session) [← Back to crosswalk](/#crosswalk) TGA Australia · Therapeutic Goods Administration # Risk-based software rules, with AI evidence rules layered on. Australia's TGA reformed its software-as-a-medical-device rules in 2021 and has since been bolting AI-specific evidence expectations onto that framework rather than writing a stand-alone AI regulation. The 2024 evidence guidance for AI-enabled software, plus the Feb 2026 refresh of the AI and medical-device software page, are now the operational reference points for any AI SaMD entering the ARTG. Posture · Developing · AI-specific evidence guidance live since 2024 Last reviewed · May 2026 Key facts Reformed software rules in force Feb 2021 Evidence guidance for AI software 2024 · updated 2026 AI regulation page last updated Feb 2026 Change posture Significant change → new inclusion §01 ## How TGA regulates AI-enabled software AI-enabled software that meets the definition of a medical device is regulated under the Therapeutic Goods (Medical Devices) Regulations 2002, as amended in 2021. Devices are classified using risk-based rules (Class I → Class III / AIMD) with software-specific classification rules (Rules 4.1–4.4 / Schedule 2) introduced in the reform. AI does not change classification logic · it raises the evidence bar, particularly around data, validation, transparency and post-market monitoring. - Most diagnostic AI lands Class IIa or IIb under the software classification rules. - Sponsors must hold a conformity assessment (TGA or comparable overseas regulator) before ARTG inclusion. - Evidence package: intended purpose, dataset, validation, human factors, post-market plan. - Australian Unique Device Identifier (AusUDI) work is rolling out and applies to software devices. §02 ## Evidence requirements for AI software The TGA evidence guidance for AI software sets out what a sponsor must demonstrate over and above the standard software evidence: clarity of intended purpose, representativeness and quality of training and test data, transparency to users about model behaviour and limitations, and a post-market plan that detects drift and adverse events. It is explicitly aligned with the IMDRF GMLP guiding principles. - Document data provenance, inclusion/exclusion criteria, and subgroup performance. - Demonstrate validation on data that is representative of the Australian deployment context. - Provide transparency information · what the model does, what it does not do, its known failure modes. - Operate a post-market monitoring plan with thresholds, escalation, and reporting. §03 ## Change control: significant change triggers re-inclusion TGA does not yet have a formal PCCP-equivalent. The default rule is that a significant change to an AI medical device · including retraining that materially alters performance or intended use · requires a new ARTG inclusion or, at minimum, a change notification to TGA. Sponsors are expected to define, in their QMS, what constitutes a significant change for their specific device. AI-specific change-control guidance is on the agency's known workplan. §04 ## Practical posture Australia is a high-leverage market: the rules are pragmatic, the comparable-overseas-regulator pathway accelerates entry for devices already cleared by FDA / EU / HC, and the evidence expectations are well-telegraphed. The risk is treating it as a copy-paste market. Validation evidence, intended purpose statements, and post-market plans need to be reworked for the Australian context, and change control needs to be planned defensively until formal PCCP guidance lands. Key takeaways 1. 01 Use the comparable-overseas-regulator pathway where possible, but rework intended purpose and evidence locally. 2. 02 Align your AI evidence package directly to the 2024/2026 TGA evidence guidance · it is explicit. 3. 03 Define a 'significant change' policy in your QMS before launch; assume material retraining triggers re-inclusion. 4. 04 Track TGA workplan updates · formal AI change-control guidance is the next likely deliverable. References - [TGA · AI and medical device software regulation (Feb 2026) ↗](https://www.tga.gov.au/products/medical-devices/software-and-artificial-intelligence-ai/manufacturing/artificial-intelligence-ai-and-medical-device-software) - [TGA · Evidence requirements for software using AI ↗](https://www.tga.gov.au/how-we-regulate/manufacturing/manufacture-medical-device/manufacture-specific-types-medical-devices/evidence-requirements-software-using-ai) - [TGA · Understanding how we regulate software-based medical devices (Feb 2026) ↗](https://www.tga.gov.au/resources/guidance/understanding-how-we-regulate-software-based-medical-devices) - [TGA · Overview of software and AI lifecycle ↗](https://www.tga.gov.au/resources/guidance/understanding-regulation-software-based-medical-devices) [ ← Previous Japan · PMDA ](/regulators/pmda)[ Next → United States · FDA ](/regulators/fda) © 2026 AI SaMD Playbook · Sponsored by Blue Goat Cyber [Crosswalk](/crosswalk)[Colophon](/colophon) Editorial reference only. Not legal or regulatory advice.