Liability for AI Harms – Submission to the UK Jurisdiction Taskforce
As AI increasingly reshapes commerce, governance and daily life, legal frameworks are being tested in ways their architects never anticipated. Stirling & Rose has submitted our response to the UK Jurisdiction Taskforce’s consultation on liability for AI harms under English private law.
Submission to the UK Jurisdiction Taskforce
For a PDF version of our submission please click here: UKJT AI Liability Consultation Submission
Key Themes and Observations
Before addressing the specific consultation questions, we set out below our observations on several key themes that arise from the draft Legal Statement and that inform our responses.
The Strength of English Common Law as a Framework
We agree with the UKJT’s foundational premise that the (English) common law, by virtue of its inherent flexibility, is broadly capable of attributing liability for AI-caused harms using existing legal principles. The incremental development of the common law through analogical reasoning has historically demonstrated its capacity to accommodate novel technologies, from the steam engine to the internet.
Having said that—and as the draft Legal Statement itself acknowledges—there remain areas of genuine uncertainty, particularly at the margins of the negligence analysis where the autonomous characteristics of AI test the boundaries of existing principles on duty of care, standard of care, and causation. We consider that the Legal Statement performs a valuable function in identifying these pressure points and articulating how the courts are likely to approach them.
We consider that, (considering AI harms) there will need to be legislative changes (e.g., with regard to the definition of (legal) person see section 3 below) and material changes to the operationalisation, enforcement and infrastructure of the law. The current analogue manifestation of the law and its administration will be entirely unfit for purpose. For example, while we vigorously agree that the “most important legal framework governing liability for loss caused by use of AI in most circumstances will be the chain of contracts that the various actors involved in the supply and use of AI have agreed…”, we consider that the technological manifestation of contracts (smart legal contracts) and new legal infrastructure to support these programmable contracts is the only appropriate real-time software instrument that can both register liability and where necessary mitigate harms or dispense enforcement on autonomous systems.
Analogue, disconnected law will be impotent without the speed and scale that matches autonomous span of control and outputs.
New infrastructure for legal guardrails will be required as technical legal brakes for embodied AI Systems (law at the edge or in the device) where the Responsible Machine Problem is most critical to solve (see section 3).
We are also happy to provide further opinion regarding our concerns in respect of the ongoing workability of the doctrine and implementation of the separation of powers. If the same AI systems are integrated into any decision making across all three arms of government, those AI systems should be technically separate (where that technical separation will become more complex as models become more complex).
The Challenge of AI Supply Chain Complexity
The draft Legal Statement’s adoption of the Ada Lovelace Institute’s supply chain taxonomy is a pragmatic and useful framing. We note, however, that the rapid evolution of the AI ecosystem—particularly the rise of agentic AI and the blurring of boundaries between developers and users (where users themselves are also developers, GPTs for example as noted in the draft Legal Statement) and the rise of agentic AI where an AI may itself be a developer of new AIs — means that any taxonomy must be understood as evolving rather than exhaustive.
In our experience advising clients on AI deployment, the practical challenge is often not identifying the correct legal principle but rather identifying which actor or actors in a complex, multi-layered supply chain bears responsibility for a given failure.
We foreshadow that as AI systems become increasingly agentic—capable of initiating actions, entering transactions, and operating with minimal (or potentially no) human oversight—the existing liability frameworks will face growing pressure. The current Legal Statement appropriately focuses on the present state of the law and AI, but we invite the UKJT in due course to consider whether supplementary guidance on agentic AI liability may be warranted, including consideration of whether AI should be granted legal personality (see section 3 below).
The International Context
We welcome the draft Legal Statement’s acknowledgment of the international context, including its alignment with the EU AI Act’s definitional approach and the OECD framework.
In our view, consistency with international norms is not merely desirable but essential for maintaining the attractiveness of English law as a governing law of choice in cross-border AI transactions. The UK’s post-Brexit position creates both an opportunity and a responsibility to demonstrate that English law remains at the forefront of providing legal certainty for emerging technologies.
1.4 The distinction between perceived and actual uncertainty
We consider the draft Legal Statement’s distinction between perceived legal uncertainty and actual legal uncertainty to be one of its most valuable contributions.
In our experience, a significant portion of the hesitancy around AI adoption in risk-averse sectors (such as financial services and healthcare) stems from perceived rather than actual legal uncertainty. By articulating that many AI liability scenarios are, in fact, amenable to straightforward analysis under existing principles, the Legal Statement serves an important educative function.
Responses to Consultation Questions
Consultation Question: Do consultees agree that the subjects that we address in the Legal Statement are appropriate and useful, and have been addressed in an appropriate and useful way? If not, what alternative issues within scope of the project do consultees think need to be addressed in this Legal Statement, or in what alternative manner would consultees wish those issues to be addressed?
2.1 Response to Question 1
Generally, Yes.
Stirling & Rose agrees that the subjects addressed in the draft Legal Statement are, appropriate and useful. In addition, we consider that the Legal Statement would benefit from additional treatment of the following matters, which fall within its stated scope:
- Agentic AI and liability allocation:
The draft Legal Statement acknowledges the rise of agentic AI in its discussion of supply chain actors (at paragraph 20) but does not explore in depth the liability implications of AI systems that autonomously initiate actions, enter transactions, or make decisions with minimal (or no) human oversight (Claude co-work for example).
We expect agentic AI to become the predominant mode of AI deployment in commercial and personal use settings. We consider that the Legal Statement should address, even if briefly, how the negligence analysis applies where the degree of human control over AI output is materially reduced or eliminated. We explore this aspect further in our response to Question 3.
- Contributory negligence in AI contexts:
The draft Legal Statement does not address the potential role of contributory negligence where a claimant’s own use or failure to use or misuse of AI contributes to the harm suffered. For example, where a non-professional user fails to review or verify AI output that subsequently causes harm, the question of apportionment between the user’s negligence and any negligent acts or omissions of the various players upstream in the supply chain is one of considerable practical importance.
We respectfully submit that the Legal Statement would be further enhanced by addressing this issue.
- Data provider liability:
The supply chain taxonomy identifies Data Providers as actors in the AI ecosystem. However, the negligence analysis does not substantively explore the circumstances in which a Data Provider might owe a duty of care for harms caused by biased, inaccurate, or incomplete data which is used to train AI.
We accept that predominantly, Data Providers furnish data pursuant to contractual arrangements, including, for example creative commons licences. However, given the critical role that training data plays in shaping AI outputs, this may warrant consideration to determine whether this aspect should also be addressed.
2.2 Response to Question 2 Maybe, with qualifications.
If the Legal Statement defines AI for the purposes of alignment of understanding, this makes sense. If the Legal Statement is trying to suggest this definition should have wider adoption (including wider legal adoption), we do not consider the chosen definition fit for purpose.
- as it includes non-falsifiable, post-rationalist assumptions in the identified limbs, and
- does not attempt to appear to contemplate the potential collapse of biology and “technology”, for example as pursued by the transhumanists and Brain Computer Interface (BCI) proponents (“technology does not occur in nature”) under its definitions.
Further, we agree the likelihood of the definition being correct is inversely proportional to the length of time that has occurred from the drafting of this Legal Statement.
We agree that a characteristics-based definition is preferable to a human-centric, teleological, or technology-specific definition for the reasons articulated in the Legal Statement. In particular, the focus on the legally salient characteristics of autonomy—comprising unpredictability of output, opacity of reasoning, and limited user control—provides a sound basis for the legal analysis that follows.
We consider the definition’s alignment with the UK Government’s White Paper on AI, the EU AI Act, and the OECD framework to be useful in so far as consistency across jurisdictions reduces the compliance burden on industry participants and enhances the utility of the Legal Statement in cross-border contexts.
Response to Question 3
Yes.
Stirling & Rose offers the following additional observations.
Liability for Increasingly Capable AI Systems of the Future
In considering the future, it is necessary to contemplate the introduction of legal personhood for AI systems and AI organisations as AI systems will commit harms that are outside the span of control or authority of legal persons.
Conclusion
Stirling & Rose commends the UKJT and its drafting team for producing a comprehensive and carefully reasoned draft Legal Statement that makes a significant contribution to legal clarity on AI liability under English private law.
Stirling & Rose would be pleased to discuss any aspect of this submission further with the UKJT and its drafting team. We are committed to supporting initiatives that enhance legal certainty for emerging technologies and that position English law as a global framework of choice for AI governance and deployment.