AI promises greater productivity, but hidden workloads, unchecked errors and growing risks are challenging that narrative. Erica Farmer explores why learning and compliance professionals must work together to rethink accountability, strengthen human oversight and equip employees with the confidence to challenge AI outputs before costly mistakes become big organisational failures.
The one thing I hear myself saying to many people right now is, ‘there is no playbook for AI’. Typically, with new system and software deployments we rely on case studies, training notes, data, previous customers and a track record of performance. However, we have encouraged people to use AI to produce more work at a faster pace with more efficiency and more effectiveness, without any of this. Therefore, it’s now becoming apparent we’ve not yet agreed what’s acceptable in AI enabled working, let alone what others should do with the work once it arrives in their in-tray.
What we miss is the extra time it takes the downstream colleague to check it
With AI, an employee can now draft guidance on a policy change in minutes. We record the time saving, equate it to department efficiencies and pat ourselves on the back. What we miss is the extra time it takes the downstream colleague to check it, decipher it and apply it to the customer context.
This is where learning and compliance teams have a real opportunity to work closer together. Everyone in the organisation needs to understand how faster production changes the demands on the rest of the process, including quality, risk and customer impact.
New work = new terms
You’ve probably heard of AI slop, which is rushed AI generated work left unchecked by a human resulting in it being poor in quality. But have you heard of botsitting yet? This refers to the unplanned, hidden labour of supervising and cleaning up AI slop; and it’s starting to rear its head in organisations.
Glean’s Work AI Index 2026 found people reporting on average 6.4 hours a week botsitting, including checking AI outputs, supplying missing context and correcting mistakes in organisational data. Alongside this, according to Skillcast’s AI in the workplace report, 68% of UK employees say AI produces inaccurate or misleading outputs at least occasionally, while only 2% said it never does. If our outputs increase in pace and decrease in quality, and the approval processes stay the same, we have quickly created new work for somebody else further down the production line, who never even saw it coming.
The evolving responsibilities for learning and compliance teams
With 25 years of experience in corporate L&D, including four in financial services and eight in a regulated environment, I expect learning and compliance teams to start to recognise these factors and help shape the changes needed.
The job is now evolving to become the conscience of the organisation; to help the business decide how work will happen, what people need to do differently, what the impact is of the change and the skills which underpin it all. This requires more than adding ‘AI training’ to a launch plan and signing off the approved AI tool. This is a new role for learning and compliance, and they must work together to be successful.
Consider a retail bank, where a team member is using a generic, open-market AI tool to turn an updated lending policy into a manager briefing. The AI summarises the standard process quickly and accurately but omits a key condition which requires a referral decision by a specialist. Without the structural guardrails and mandatory human review built into dedicated compliance tools, moments such as this blind spot would easily go unnoticed. Instead, under a time constraint due to a productivity target, the team member simply waves it through, resulting in a high-risk customer being lent too much money, unable to pay it back.
It’s not morally correct to blame the employee, right? They’ve done what’s been asked of them within the guidelines they’ve received. They are undertaking the tasks in their job description and using the AI tools as they have been told to.
So, what needs to change? Oversight and ownership. Often, we are too quick to accept confidently wrong answers from AI, and too slow to become aware of the consequences. This looks like:
- Learning opportunities in context of true examples, including role relevance and real-life compliance data. People working through real situations, comparing real outputs with the approved policies and explaining when they would seek advice, not just working through a list of prompts of use cases which are unrelated to their day to day
- Everyone’s accountable for their own AI outputs and can justify their reasoning. People can confidently provide the context that a reviewer needs, including the source and what has already been checked
- Confidence to challenge others, no matter their position in the hierarchy. Leaders praise those who identify incorrect AI use and actively work to make things better. What a fantastic Compliance performance indicator this would be!
- Compliance teams examining data which they would have typically dismissed, including work in progress, instances of repeat work and organisational AI capability scores, to undercover risks which are in progress and not breaches after they occur
The human in the loop is more important than ever
The FCA’s October 2026 good practice publication emphasises that human oversight depends on appropriate authority, information, capability and escalation routes. This means three things in the context of this article:
- Leaders identifying who will review the work, without them becoming begrudging botsitters, ensuring they have the authority to stop any issues in their tracks without reprimand
- In partnership with compliance teams, learning teams being set up so they can update learning materials with agility and confidence, distributing to those in need without compromising on quality or accuracy
- In addition to this, where there are many new AI start-ups in the market, particularly in learning and HR, when selecting suppliers organisations should prioritise working with trusted partners, who can demonstrate a track record of effectiveness and compliance. Confidence in relationships and capability in this area are worth their weight in gold
Transformation relies on successful partnerships
The 2026 Gartner report, ‘Dynamic Redesign: The Next Evolution of Work Transformation’, describes the need of AI enablement to become continuous work design for sustainable AI transformation. In short this means we need to continually move forward to achieve this.
The great news is that compliance and learning have a significant opportunity to lead this together: spotting where new ways of working create risk, challenging practices that no longer fit and equipping people to act before problems take hold, not just rushing to get finished at the end of the day.
Erica Farmer is AI and Future of Work Speaker, Trainer and Co-founder at Quantum Rise and author of AI For People Professionals
Erica is also keynote speaker at the TJ Conference 2027
If this article piqued your interest, you’ll probably find value in our conference!
Find out about keynote speaker Erica Farmer, her opening session on the
AI Fire Hydrant and what it means for productivity, performance and L&D
Don’t miss 50% off your ticket for the first 30 only!
