Technology changes. Professional duties do not
The trend of lawyers getting caught out by AI hallucinations continues to unfold with the predictability of an old slapstick silent film. You can see the disaster happening in a way that the main characters seemingly cannot. Like a man carrying a plank of wood and not being aware that, as he turns around, the plank he is holding will hit someone else on the head.
A global pandemic (of professional negligence)
Pinsent Masons. Sullivan & Cromwell. Various international firms and chambers. The examples span multiple jurisdictions and practice areas. No one appears entirely immune.
I used to maintain a list of the firms involved until I happened upon Damien Charlotin’s AI Hallucination Cases Database, which now tracks in excess of 1,400 cases globally involving hallucinated authorities or AI-generated inaccuracies in litigation and arbitration proceedings.
The dominant media narrative tends to frame these incidents as a shocking new technological risk: AI generating plausible sounding but ultimately fictitious case citations, authorities or quotations which then find their way into legal submissions and client work.
To me, however, this is not a problem of rogue sentient AI causing mischief.
Rather, it is a supervision and workflow problem.
Supervising Work
When I was a newly qualified lawyer, a supervising partner once told me regarding document review:
“You review it and review it and review it until your eyes bleed. Then you review it again.”
At the global law firms where I practised, drafts would typically then undergo further review by the matter partner before going out of the door.
The underlying professional obligation was obvious: the human lawyer remained responsible for the work product. That remains consistent with lawyers’ duties of care and regulatory guidance across multiple jurisdictions where I operate, including Singapore.
AI changes the speed and volume at which errors can be generated. It does not remove the obligation for adequate supervision, verification and accountability for professional standards.
In many of the recent hallucination cases, the real issue appears not to be the technology itself, but combinations of:
- over-trust;
- lack of verification;
- limited or no supervision;
- time pressure;
- workflow failure; and
- professional complacency.
In other words, very old professional risks wearing modern clothes.
The more important issue is therefore not whether AI occasionally fabricates authorities. We already know that it does.
The real question is whether professional institutions possess the supervision culture, scepticism and internal discipline necessary to ensure that human judgment is meaningfully engaged with the work product.
Good supervision has always produced commercial and professional benefits far beyond simple error reduction:
- better quality work;
- lower write-offs and rework;
- stronger delegation and leverage;
- faster skills development;
- higher staff retention; and
- greater client confidence.
Yet many professional organisations still treat supervision as an optional aspiration rather than an enforced professional standard.
The irony is that the underlying supervisory disciplines themselves are not especially exotic.
When you give somebody a piece of work, make sure they know what they’re doing. If you’ve got more than one person on the job, make sure the left hand knows what the right hand is up to. Proofread the document and verify the citations and authorities before the draft is finalised.
This isn’t advanced PhD stuff — even an insurance lawyer like me can do it.
AI does not remove the need for those disciplines.
If anything, it increases their importance dramatically.
Technology is potentially wonderful, but professional services firms will never fully capture its benefits if they lack the discipline to do the basics properly.
