Research & RankingsSurvey report and team interviewsTechnology
Firm Declares Shadow AI Rare After 96.9% of Staff Decline to Discuss It
Wetherby Dodd welcomes a reassuring absence of answers, while three unauthorised teams offer to improve the research.
By Beatrice Vale · · 3 min read
Wetherby Dodd has declared unapproved AI use “limited and well understood” after 96.9 per cent of staff left its usage question unanswered.
Admissions rose from 2.1 per cent to 3.1 per cent, an increase of one percentage point. Management called this “a modest movement within an overwhelmingly compliant population”.
The population had not said it was compliant. It had said nothing.
The firm’s research summary said: “We found very little evidence of activity among the people who did not supply evidence.”
The findings were presented to partners in a green pie chart. The unanswered portion was labelled Established Practice.
Methods note
Both percentages use all staff invited to answer as the denominator. In the current survey, 3.1 per cent admitted using unapproved AI and 96.9 per cent provided no answer. The latter figure contains no recorded denials of use.
The questionnaire requested staff numbers “to ensure anonymity can be monitored”. Its introduction promised a non-judgemental discussion of any admitted policy breaches, with the relevant line manager copied in for support.
A methods appendix recommended further work to understand non-response. The executive summary recommended congratulating staff.
Team one: Document Throughput
A subsequent review found a team using an unapproved tool to summarise documents for fee-earners who had become impatient with the approved tool’s procurement timetable.
Its lead described this as “an interim arrangement”.
“We’re filling a gap until the firm supplies the capability,” he said. “Shadow AI is when people set up parallel systems without permission. Knowledge Continuity does that. So does the Client Response Group.”
Asked who authorised his own system, he produced seventeen partner emails asking whether their summaries were ready.
He has offered to automate the next survey. His proposal asks management not to circulate it until the sponsorship question is settled.
Team two: Knowledge Continuity
The second team has built a searchable collection of precedents using a service that does not appear on the firm’s technology register.
“That register is out of date,” its lead said. “For example, we aren’t on it.”
She called the project an interim arrangement pending formal approval. Document Throughput, by contrast, was “a completely unauthorised operation”, while the Client Response Group was “unauthorised and without even a terms-of-reference document”.
Knowledge Continuity’s terms of reference are awaiting approval.
The team has separately volunteered to rebuild the survey. It requested a confidential conversation because involving the other two groups would “normalise shadow provision”.
Team three: Client Response
The third team uses an unapproved assistant to prepare first drafts of routine client updates.
Its supervising partner said the project was an interim arrangement and therefore did not yet constitute a system. He regarded both other teams as unauthorised.
“You cannot just decide something is approved because it would be useful,” he said.
He then explained that his arrangement had to continue because clients found it useful.
His team has submitted its own offer to create a more reliable survey, marked FOR YOUR EYES ONLY. None of the three teams knows the other two have also volunteered.
Management has welcomed the unexpected availability of three experienced internal delivery teams. Each has been asked to demonstrate its existing capabilities before any decision is taken.
All three have requested that the demonstration be excluded from the technology register until the pilot is complete.
Something worth not billing for.
Get The Unbillable every week.
Email delivery is coming soon. Until then, The Brief is published here every week.
Read the latest BriefRelated stories

Artificial Intelligence · News report and incident log
Partner Blames AI Hallucination for LinkedIn Post Dictated While Hallucinating
An emergency review has found the software’s only contribution was a correctly placed comma.
By Celia Firth, Partnership Correspondent17 Oct 2026 · 3 min read

Disputes · Reconstructed rolling report
Opposing Counsel’s AI Agents Launch Swarm Attacks Before Either Partner Has Read the Email
Autonomous dispute resolution achieves the first fully automated collapse in professional relations.
By Celia Firth, Partnership Correspondent17 Oct 2026 · 3 min read

Innovation Interview · Interview and demonstration
Firm Replaces Every Desk with VR Glasses Offering Views of Places Staff Cannot Visit
Hargreaves Crake’s new Head of Innovation demonstrates flexibility in both directions, neither of which leads home.
By Martin Pavey17 Oct 2026 · 3 min read