12,000 replies, zero liability rule: this week in AI

This week's throughline is a gap between how much work has gone into governing healthcare AI and how little of it has actually landed as a rule a practice can follow. A 119-page UK report drawing on 12,000 responses came out on Thursday and still left the hardest question open. A promised UK data-protection guidance is now several months late. Three US lawsuits over AI scribes are still just sitting there, unresolved. None of that is a reason to wait before using AI. It's a reason to stop assuming a finished rulebook is coming to tell you what's allowed, and start documenting what your own practice does instead.

1. The UK's biggest ever AI healthcare consultation still hasn't answered who's liable

The National Commission into the Regulation of AI in Healthcare, established by the MHRA in September 2025 and chaired by NHS clinicians Professors Alastair Denniston and Henrietta Hughes, published its recommendations on 10 September 2026 after gathering evidence from more than 12,000 people, the largest engagement exercise ever undertaken in the UK on regulating healthcare technology. The report's Recommendation 32 says contracts between AI device manufacturers and healthcare providers should explicitly allocate responsibility for controlling risk, because current negligence procedures tend to direct claims at clinicians and providers, who carry the clearest duty of care, regardless of where an AI system itself went wrong. The Royal College of Pharmacy and the Pharmacists' Defence Association both welcomed the finding but said it confirms a problem they had already flagged rather than resolving it. A cross-government response setting out what will actually change is still to come. Source: GOV.UK and the Medicines and Healthcare products Regulatory Agency, 10 September 2026.

The number: 12,000. The response count behind the UK's largest healthcare-technology consultation to date, and still no finished answer on who is liable when a healthcare AI tool causes harm.

The so-what: until a framework exists that reassigns it, liability for an AI tool's failure sits with the practice and the clinician using it, by default, under ordinary negligence and professional-conduct rules. The one thing an independent practice can do now, ahead of any national framework, is make sure its own contract with each AI vendor says something about who is responsible for what, rather than leaving that question to be answered for the first time in a dispute. That is exactly the same gap this site found dental practices sitting in in GDC's safeguarding guidance predates AI: multiple documents exist, none of them assign responsibility for the specific case, and waiting for a regulator to close that gap is not a plan.

2. The ICO's promised summer guidance on automated health decisions is now overdue

The ICO's consultation on draft guidance covering the Data (Use and Access) Act 2025's changes to automated decision-making closed on 29 May 2026, with final guidance originally expected in summer 2026. As of this week it has not appeared, and the ICO's own published guidance pages on AI and automated decisions still describe the law as it stood before the 5 February 2026 reforms took effect. Source: Information Commissioner's Office consultation page, checked September 2026.

The number: zero. The number of finished, updated ICO guidance documents currently available reflecting a law that has already been in force for more than seven months.

The so-what: the delay doesn't create a gap in the law itself, only in the regulator's explanation of it. Article 22B's restriction on solely automated decisions using special category data, which covers most AI-scored mental health outcome measures, is already in force regardless of whether the ICO's guidance has caught up. We set out exactly what that means in practice in Article 22 loosened. Not for health data., and nothing this week changes that read. Don't wait for the ICO's guidance to catch up before applying the statute that's already binding.

3. Three AI scribe consent lawsuits are still open, with no ruling in sight

Sharp HealthCare, sued first in San Diego Superior Court over its use of Abridge's ambient AI scribe, and Sutter Health and MemorialCare, named in a related class action filed in the US District Court for the Northern District of California in April 2026, all remain in active litigation with no ruling on the merits identified as of this week. The claims allege patients were recorded without consent and, in Sharp's case, that the AI tool auto-inserted consent statements into charts that patients say never happened. None of the three complaints allege a HIPAA violation; all rely on state wiretap and consent law instead. Source: court filings summarised by TechTarget, Becker's Hospital Review and Paubox, 2026.

The number: three. Health systems now named in AI scribe consent litigation, with the count growing and no dismissal or settlement yet on any of them.

The so-what: a health system's own scale doesn't insulate it here, and neither would a signed BAA. These claims work precisely because HIPAA compliance and state consent law are two separate legal tests, a point we've made before in your AI scribe's BAA covers HIPAA, it doesn't cover this $5,000-a-patient lawsuit. If your practice records sessions with an AI scribe in an all-party consent state, the open status of these cases is not a reason to relax, it's a reason to check your own consent workflow now rather than after a claim is filed against you.

4. APTA's 8.2% shortfall figure is drawing confirmation, not pushback

APTA's own microsimulation forecast, projecting a physical therapist shortfall rising from 5.2% in 2022 to 8.2% by 2027 before easing to 3.3% by 2037, continues to be corroborated by workforce survey data rather than disputed. Nearly three-quarters of practising physical therapists report being at or over capacity, one in four say they have had to turn patients away, and administrative burden alongside low reimbursement is cited as a leading driver pushing clinicians out of the profession, with 8.9% of skilled-nursing PTs alone planning to leave or retire within the year. Source: American Physical Therapy Association workforce forecast and member survey data, 2025 to 2026.

The number: 8.9%. The share of physical therapists in skilled nursing settings alone who say they plan to leave or retire in the next year, on top of the national shortfall already forecast.

The so-what: the reception this figure is getting matters more than the figure itself. Nobody in the profession is arguing the shortage isn't real, they're arguing about how fast it gets worse. We covered the underlying data and what it means for a single physio practice's own admin load in 8.2% by 2027: the physical therapist shortfall. This week's reporting adds one detail worth acting on directly: administrative burden, not clinical burnout alone, is the named reason clinicians give for leaving, which makes it the one lever an individual practice can pull without waiting for a single new graduate to enter the workforce.

5. A new argument says AI could grow the clinical workforce, not shrink it

In a Perspective published in the New England Journal of Medicine on 12 September 2026, Dr Dhruv Khullar of Weill Cornell Medicine argued against the assumption that AI agents will mean fewer clinical jobs. He pointed to cataract surgery and joint replacement as precedents: both became more efficient through technology, and demand for both grew rather than shrank as a result, a pattern economists call Jevons paradox. He also challenged the "lump of labor" assumption that there is a fixed amount of clinical work to automate away, arguing AI may instead create new treatments and specialisations that didn't previously exist. Source: The New England Journal of Medicine, 12 September 2026, via Weill Cornell Medicine.

The number: two. The named historical precedents (cataract surgery, joint replacement) the argument rests on, not a workforce study of AI itself.

The so-what: treat this as a reasoned argument, not a settled forecast. It doesn't prove AI will grow headcount in any specific specialty, and it isn't a reason to defer a staffing decision either way. What it does usefully challenge is the fear, common among owners weighing AI adoption, that every task an AI tool takes on is a role that disappears. For an independent practice, the more realistic near-term effect is the one this site has documented specialty by specialty all year: AI absorbs admin load so existing staff can see more patients, not that it replaces the staff themselves.

What this week adds up to

Every one of this week's five stories is unfinished. A national liability framework, a data-protection guidance document, three lawsuits, and even an academic argument about the workforce's future are all still open questions rather than closed ones. That's the normal state of AI regulation and litigation right now, not a reason to treat any of it as settled in either direction. What an independent practice controls in the meantime is smaller and more concrete: what its own contracts say about responsibility, what its own consent workflow actually asks of patients, and where its own admin burden sits before deciding what AI should be asked to fix first.

If you want a structured, independent read on where your own practice sits against those questions, the AI Opportunity and Growth Assessment benchmarks you against the CARE Framework in two weeks. Or book a 20-minute discovery call and we will work through what this week's stories mean for the tools you already run.