The Law Society has published a significant report on the future of agentic AI in the justice system, examining its implications for courts, tribunals, and alternative dispute resolution.
It seems like only yesterday that the Law Society was publishing its guidance on the use of LLMs and warning of the risks associated with using generative AI tools.
Now, like many organisations, the speed of technological developments has meant that its attention has firmly turned to agentic AI, something that that the Law Society states “represents something different”.
The report explains the significance of this shift in stark terms: “for the first time, we are moving from machines that help people make decisions towards systems that can pursue goals, take actions, coordinate tasks and interact with other systems with increasing autonomy. The question is no longer simply what information technology can provide. It is what actions we are prepared to allow technology to take. That distinction matters”.
Given the stakes, the Report is unsurprisingly rather lengthy, touching upon everything from how agentic AI might improve access to justice and assisting litigants, to its use in the administration of courts and tribunals, and the various risks associated with its deployment.
Drawing on discussions with practitioners from a range of disciplines, the report identifies a broad consensus that the routine deployment of agentic AI in the justice system in England and Wales will require a more robust regulatory framework than that applicable to conventional generative AI.
Quite what form that framework should take remains a matter of disagreement. Some favour building on existing regulatory arrangements, while others advocate an approach similar to that adopted under the EU AI Act. But, whatever the model, the report identifies four ‘regulatory considerations’ that are specific to agentic AI and go beyond merely adjusting existing rules:
- Regulators will need to establish a working definition of agentic AI;
- Deploying agentic AI over a simpler alternative should require explicit justification;
- The “autonomous, multi-step nature of agentic AI requires a higher regulatory threshold” than applied to other AI tools; and
- Agentic AI systems used in the context of the justice system “should be limited to defined, verified sources rather than being free to draw on unconstrained material”.
The Report also emphasises the need to resolve questions about who bears responsibility when things go wrong, and whether human oversight will remain an adequate safeguard as agentic AI becomes increasingly capable.
To read the report in full, click here.