The Work Agenda 2026 podcast series: Episode 02 – The inside job: tackling insider threats and AI-driven data risks

Technical defences alone won’t stop misuse, error or rogue insiders. From careless clicks to malicious insiders, employees remain the most likely source of data compromise.

We unpack how everyday access rights and behaviours, including misuse of AI systems, become high impact breaches/risk incidents, and what “appropriate” security measures really mean in practice. Expect clear guidance on what good governance means, risk assessment, when and how to notify data regulators (around the world) and data subjects, and how to close incidents well to reduce repeat risk.

This episode is part of a mini-series covering the topics discussed at The Work Agenda 2026 conference, we hope you enjoy listening. Remember to share with colleagues and if you do have any comments, we would love to hear them.

Transcript

I'm Lucy Lewis and thanks for tuning in to one of our live podcast recordings from The Work Agenda 2026. We know technical measures alone can't guarantee protection against misuse, accidental errors or insider threats, and so employees remain the most likely source of data compromise. In one of our breakout sessions, Lewis Silkin partners, Alex Milner-Smith and Bryony Long unpacked how everyday access rights and behaviors, including things like misuse of AI systems, become high impact incidents. We also discussed what appropriate security measures mean, when and how to notify regulators and data subjects, and how to close instance well to minimize future risk.

Hi and good afternoon, everyone. I'm Bryony Long, I co-head the Data and Privacy Group at Lewis-Ilkin, along with my other co-head over here, Alex Milner-Smith. ⁓ Alex and I are here to talk to you today specifically on two topics that we are dealing with day in, day out, ⁓ involving workplace data. ⁓ And the first one is ⁓ tackling insider threats and AI-driven data risks.

Now, ⁓ I guess both of these topics have always been, well, the first one, tackling insider threats, has been an issue for a long time. And I suspect a lot of you are dealing with erroneous stuff, downloading data, putting it places they shouldn't be, and dealing with the consequences of that. But what we're beginning to see a real rise of now is use of AI within the workplace that isn't possibly authorised, people putting data into AI tools, and then doing all sorts of things which may or may not be good for the business. We're finding that we are dealing with and advising clients a lot on how to manage that, how to deal with it when it happens, but also how to be ready for it and prevent it from happening in the first place. So what we thought we would do was rather than talk at you, well, we probably will talk at you for quite a bit, but what we want to do is do it in a slightly different way in that we want to create a case study built off the case studies that I think you've been seeing in the plenary sessions, then we're going to of examine some of the issues together. 

So this is our example. Max is asked by the HR team to run a report on this year's bonus information.

He decides to download, so he's actually been asked by HR, so Max is from the IT department. He's actually been asked by HR to run a report on the bonus information. He downloads all the information that he's authorized to access, but then he decides to put this information into ChatGPT on his personal phone.
He knows that he's done a similar thing before for personal reasons, and he thinks that this way of using ChatGPT will get a better result for him so that ultimately the data that he feeds back will be better and more useful for the business.

In particular, he prompts ChatGPT to see if they can identify any patterns, and in particular, who's received the highest bonus and which sections of the staff have received highest bonus. However, also thinks in his worlds of wisdom that he can improve his results if he adds some additional appraisal information. So then Max decides, again, because he's got access to this, that he's going to include names, roles, ratings, narrative comments, and inputs this data into ChatGPT too. That is the scenario. I guess what we this, unfortunately, has happened. Clients have come to us and said this has happened. I guess the first thing is over to you, Alex. What would you be saying to the client right now?

Yeah, I mean, so the first thing to note is there's myriad different issues here. Is it a data breach? We'll talk about that in a second. I won't talk about it too much because we're going to talk about that with the insider threat scenario in a second. But there's then what are the failures of AI governance that have happened here? We'll talk about potentially what we need to do going forward to stop this happening. There are then lots of employment issues that neither Bryony nor I are as capable of talking about, we will talk in generalities about them. Is there a breach of a disciplinary policy here? What action can you take? But before I run through some of those, it is very important to note in this scenario, he has access to this kind of stuff. It's a perfectly authorised access.

We will see later, maybe we need to think about the access within businesses, because it does surprise us when we deal with these situations quite how wide access is, certainly within the IT team. I’m not casting aspersions about IT teams, very venerable people who in the main are honest. But occasionally there have been some things that we've dealt with at Lewis Silkin, big fraud cases that have emerged from similar people who have enormous access to data. So that idea of custodian limits and so on is just good data loss prevention. We'll talk about that in a second. The second point, there isn't any malice here at all. Max, I think he was initially called Alex at one point, but that was too close to the bone and I luckily got through that issue. But there is no malice from him here at all. He does think he is doing what is good for the business. But I think what he, despite being in IT and no doubt very expert in these things, the risk profile that he's potentially producing here is enormous. So if we start with the breach point, Bryony, I think some people would think, this is totally authorised. There is no breach here. But, just in layperson's terms, there is a misuse of systems here. This is exactly the same as potentially leaving a file of documents on a train, giving a file of documents to a random consultant that is not approved, and slightly less and newer money, sending a huge swathe of documents to your Gmail or Hotmail or Outlook account. We would all immediately think that is a breach of policy. And that is what has happened here. 

And if we think of Article 412, GDPR UK GDPR that the definition of a breach it talks about a breach of security now because I'm a mildly aggressive lawyer as is Bryony and there are arguments about whether it is a true breach of security, but I think our European continental European friends and partners would probably regard this as a breach of security that potentially leads to unauthorised disclosure unauthorised access who knows. So we could in this scenario be in a reporting situation, whether it's to a regulator, to data subjects. I mean, it's the whole of this company. It could be 3 people. It could be 3,000 people. It could be 30,000 people. So from a breach perspective, I'm not going to run through that now because we'll do it, think, Bryony, in the later scenario. But this probably is a data breach for me.

The things that are triggered from that depend on your organisation, but no doubt you've got a standing breach notification protocol. You probably have to speak to multiple people around the world. You probably have to notify your insurers. So this is an enormously serious issue. The second point, just from a risk perspective, what are we actually talking about here? And why has there been a failure of governance, probably? 

I have ChatGPT on my phone. Those who have seen me speak before only ever have two stories. I'm going to repeat the other story. I use it to create stories and images for my children to try and improve my creativity when being a parent. But I don't write their names into that. I don't give any personal data. But it's a B2C operation. I know that I have signed up to the most aggressive terms going and that OpenAI are going to store my prompts and are going to use my prompts for training. And were I foolish enough, as I do with my own Images, foolish enough to upload my own photo, I know it is going into the ether and some of you may be lucky enough for it to appear on your ChatGPTs in some various form. But the difference is when we are at an enterprise level with whatever tools you use in your businesses, be it Copilot or you're using maybe ChatGPT, the enterprise level versions you could have, or even the API versions, or even higher. Even at the base level before negotiation, it is so much more protective of an enterprise. And when you start negotiating it, I'm happy to say we at Lewis Silkin have negotiated with some of our vendors. I won't talk about specific vendors, but to absolutely be clear that no prompts are stored, no data we put in will be used for training. It is an absolutely segregated sort of unit.

That is not happening here. This person has essentially put all the bonus information and all the appraisal information, pretty much the crown jewels of confidential information within the HR world, save perhaps for medical data, into a consumer level app. 

There are also other things. It shares data with the US, his could have a European construct in it as well, and it could breach all kind of rules. So the fact is that is the risk that we're looking at. ChatGPT could have its own breach. This data could then reemerge in someone else's outputs. The risk is enormous, and you have absolutely no control of it now.

So that should be obvious to everyone. The more interesting point is what could we have done, perhaps in an employment or HR or DLP world for us, to prevent this. Obviously, training, everyone says training. We're going to talk about AI literacy later. But that is just still giving the risk to a human. And a human could still make this decision no matter what training you have.

There needs to be some kind of system within your data loss prevention posture to track this, to prevent anyone downloading these apps, maybe onto their company phones. We'll talk about the difference between BYOD and CYOD in a second. But equally, it can go further. There are systems now, just like there are systems that flag if any huge amount of data is sent to Gmail. I won't say our IT teams, IT teams should know that immediately and action can be taken.

Now, where it gets difficult is it's on his personal device. But there are ways to stop that by having, even if someone is using a personal device, there should be a BYOD policy that makes sure that any company data is within a mobile device management wrapper, so no data can just be dragged from one system to another without us knowing. And the number of businesses, Bryony, that you know, this just hasn't been thought about in this new world. It's quite surprising to us because this is the work of a moment and it could really be, maybe nothing happens obviously, but it could be an enormously serious risk. 

So that isn't actually how I would advise clients in some kind of enormous chunk of rubbish. It would be more of a dialogue, but they're the kind of thoughts that I've been having about this Bryony, say. I don't know what you think.

Obviously, I agree with all of your thoughts. I think one of the things that I often talk to clients about is, when we are implementing AI solutions for the workplace, which we are, lots of clients are using various different tools, for things like summarisation or analytics or that sort of thing, you would always have, or we would always advise, least having do's and don'ts, very simple policies that are very, very clear for how to use a particular tool. You can have your overarching policy, which will say only use authorised systems, but actually having specific do's and don'ts which are tailored for the limitations of that particular tool is really, important to make sure, A the tool is being used as it should be, but also that you're getting the results. Because every single tool that hopefully is being deployed within the workplace has gone through some form of risk assessment. And so you will be comfortable, or at least the business should be comfortable, that the tool is going to operate as it says it will.

Obviously, we can’t completely get rid of hallucinations when it comes to generative AI, i.e. making up inaccurate results and that sort of thing. But at least people are aware that the tool might hallucinate, they'll be able to make sure that they check the source material, they won't over rely on the tool because they know it's AI. Whereas if you are using tools that haven't been through that process, you don't have clear do's and don'ts about how that tool is going to be used. That is also, for me, a massive red flag around using that tool. And this is exactly the example here with ChatGPT. Now ChatGPT can do million different things, but it has its limitations. And this particular ask of it probably is not beyond its ability to do a relatively decent job. But there are certain things that you ask it to do where it can produce absolutely, well rubbish, to be honest. Probably not in your artwork that you do with your children. But I know some certain questions I've asked have come up with all sorts of weird and wonderful ideas. And then when you've got staff relying on some of that information and then using that information to make decisions, I hasten to add, or anything like that, then you've got some much bigger problems. So that is also one of the other things that I would always just be mindful of.

And another point that comes up many times with clients, some clients have gone so far in terms of their AI governance that their employees feel restricted, even in terms of tools that have been approved. 

There are 40, 50 page AI governance or AI use policies. I'm sure you have these in various forms. But you are so restricted that it sort of forces people to use this. 
The best clients have approved tools and then approved use cases, but that aren't so narrow that it means people feel that they can't explore the way they're using AI. In this case, could be, I mean, it's very difficult because an HR example, and I'm afraid our advice with HR would always be to take a step back and actually go to legal and get approval. But just for the sake of heuristic here, perhaps there is an approved large language model at this business where certain HR statistics or analytics use cases have been approved as long as perhaps you de-anonymise the data set before you put it in. And if you've thought about that before, gives employees the agency to use the systems that are approved and have all your back-end protection without having to do this kind of rubbish. But as we know, Bryony, and people in the room, obviously, HR, very senior HR people and employment lawyers, AI in businesses is not just the preserve of one team. So you might want to do that, but you probably do need to talk to your data teams, to your AI governance teams, to your risk teams. All businesses are different. We have found that as well, that totally siloed approaches mean that some of these ideas are missed. I mean, that's really all I've got to say on that. But no one's got a question, so it must all be perfect.

Well, the other thing I guess is also to add, is, and I'm going to talk very high level of this, Alex, you might be able to talk a bit more deeper on this, but there is obviously a risk of claim as well. I mean, exactly what damage someone has suffered or what loss someone has suffered is quite difficult to articulate, particularly from a data protection perspective. But certainly, you can say loss of control of your data, which is a kind of valid heads of loss. And so you've also got to think about how to protect yourself in that scenario.

One thing that, for example, if we were using an AI tool that was authorised to to help do analytics on bonus information, one of the things that we would be advising an organization to do is in your privacy notice, be updating the staff that their data will be used for analytics purposes and actually giving them the rights to opt out. Now, not everybody does that, but I think technically for that kind of purpose, staff possibly should have the right to opt out of their data being analysed for that purpose.

But all of those sorts of things won't have happened here because this tool hasn't gone through the approval process, so potentially there are potentially employment claims. I'm not going to start going into that, but there's certainly some data claims there that could be potentially raised in a grievance.

Yeah, I’ll make three points on that. From employment claims, neither Bryony or I could speak too much of that. I mean, this could, you know, if someone the fact is, if people found out about this, it could be, it could found a breach of trust and confidence claim. There's all kinds of things that happening here.  

From a data claim perspective, data claims will also potentially arise from the next scenario but I'll say this now. I was going to say it later, but I'll say it now. In the UK, less so across continental Europe, because they've always been less litigious, middle ground Republic of Ireland, it is changing but hopefully they're 10 years away from getting to us and we're 10 years away from America. But we are seeing, I'm afraid, more and more low value data claims. And a lot of them come from the HR space. And although we were very hopeful that the Court of Appeal in the paymaster decision would uphold the fact that there should be a de minimis threshold for these claims. I'm afraid the Court of Appeal said there is no de minimis threshold. It's going to the Supreme Court. But it does mean, Bryony dealt with one in Ireland recently. Our team are dealing with a number of them. There are claims that are not worth their salt. You probably don't see them in the employment world. But they are going, maybe they're not going to the High Court, but they're going to the County Court.

They're complex claims because this is not fun stuff to deal with. It's underpinned by a lot of guidance and legislation and a lot of case law. It could cost you 30,000, 40,000 pounds to defend. And yet the claim is only worth 500 quid. One of those you can deal with, but if you become a soft touch and you're dealing with 50 a year. So, I'm afraid that you're just going to have to get used to that. So that just adds to the risk. And then the final point, just in relation to Max, if you are going to have policies around AI use and so on, need to be back to back with your disciplinary policies. In our ivory data tower, we often reference this and never see what happens. But it would be disappointing that if somehow someone got away with this not as a disciplinary offence. Because it is just a standard old money you have not followed whatever policies you have in place, and you've put the business into material risk. And in my opinion, it's a gross misconduct offence, which might be a good way of moving on to the next slide.

The other thing just to also, I don’t want to scaremonger, but it is actually a marketing tactic that's sort of promoting what we call shadow AI to people within the organization so that they do start using it giving them a free license saying all this will really help you improve your role and you can do better at work and then they'll start using it with and the marketing tactic which is an old marketing that's not just deliberately about AI as an old IT tactic but would be the idea is it will encourage use at kind of a level of the business and then they'll suddenly say, this is really brilliant and it becomes reliant and then they'll push it up to more senior people in the business who actually can make business decisions. And that is, as I say, something that, you know, protecting against that when you've got aggressive marketeers at you trying to help you improve your job is quite difficult.

I mean, anecdotally, we receive those emails every week, these new AI products that we, sound incredible, they never work, but we would love to use them, but luckily we're constrained, because otherwise I'd be using all these kind of things. And in the HR world, there are so many, whether it's recruitment sifting tools, which is the standard one that everyone talks about, or we're moving into use of AI and performance. If an HR operative was somehow able to just start using that, you have the risk of the security risk, but as Bryony was intimating earlier, you have the terrible inaccuracy risk as well. It hasn't been tested. What is the output that's happening here?

Right. So moving on to the next slide.

I haven't been as clever as Bryony. I'm sorry, I'm have to turn around and twist my neck. So we didn't sack him.

And those of you who have a keen eye will see this scenario is obviously based on the Morrisons case. But it is still a very, very apt scenario that we deal with almost every week. So following being reprimanded for using an unauthorized AI tool, he only got an oral warning. Should we do that? I don't know. He felt very aggrieved, as one does. He felt he'd used his initiative and was delivering what was asked of him in an innovative and efficient way. Also, given his IT expertise, he felt he'd weighed up the risks and that the company should be adopting this tech with this as the perfect use case. As a senior member of the tech team, had enterprise-wide access to all systems, including HR and payroll and every other system under the sun. Fuelled by anger and upset, in a moment of retaliation, Max decided to download all the organisation's payroll data and upload it to the dark web. So a ridiculous scenario, but what happened in the Morrisons case. He had been aggrieved by a disciplinary process about a bag of white powder. It turned out I think he was just selling protein powder or something like that, but he was aggrieved by that. The system access wasn't changed and he uploaded 100,000 payroll data sets. He went to prison, by the way, but we'll talk about that. So, Bryony, what are the key issues here? What does a client need to know? What have they done wrong? What have they done right?

I guess Alex has given it away. If you've read the Morrisons case, then you probably know the answer to this question. But anyway, we will just trot through. First thing is that you would be thinking about, is this a personal data breach? Has there been a breach of security that's led to unauthorised access or disclosure of data? I think clearly there has been. Even though he had access to the data in the first place, query if he should have had access to all of that data, but even if he has authorised access to that data, he's clearly used it in an unauthorised manner, so that would be considered a breach of security. So you're definitely into personal data breach territory.

Then when you're sort of looking at sort of things like whether or not you need to notify, the first thing you need to be thinking about is what are the harms that can be caused by this. Now, I mean, clearly there's lots of harm here. There's potential fraud identity. There's harm for the business. There's harm from a potential discrimination. There's a professional secrecy. There's lots of things. I mean, the main one from an individual's perspective is the fact that all of their very personal information is out on a deep dark web and can be used for fraud. But still, there are quite a few harms that you could say would follow from that, not just as a safe for the individual, but for the business as well. Obviously, when you're looking at it from a personal data perspective, you're only really looking at it from the individual's perspective, but still. And then you need to think about, right, well, OK, it's harm. There is a potential of harm. How can it be mitigated? 

And unfortunately, when the data is gone, into the deep dark web, there ain't much you can do to bring it back. Obviously, if he'd overshared it with a colleague, then obviously you could say to the colleague, this was shared in error, please delete it and don't forward it on. And you can take some measures. So just because you've unlawfully disclosed data doesn't necessarily mean you can't bring it back. But in this case, the web, it's gone. There's not much you can do there. There are certain things like credit monitoring that you can do to try and protect the individuals.

So that would be monitoring their data that's used in fraud. But again, that's not really going to get you to a level where you're comfortable there isn't a risk of harm. So you're then at the stage where, right, well, what can you, what do you need to do? Who do we need to notify?

First of all, if there is a risk of harm, you are obliged under data protection legislation to notify your regulator. So in this case, it would be the ICO, and you have to do that within 72 hours. Now, unfortunately, we do quite a lot of notifications.

I'm not going to say 72 hours is the be all and end all. If you don't do it in 72 hours, I don't think you're going to get massively criticised, but you should get as close to 72 hours as you possibly can, which is why, unfortunately, Friday nights are never relaxing for Alex and I, because data incidents always happen on Friday evenings. And we've always got to do something within 72 hours to report it. 

Sometimes the easy ones, actually, the ones that actually act fine, can just report it straight away. But anyway, it's likely that you would have to report it. It is also likely in this case, would have to, it would be likely to result in given the level of detail, the data that's gone, it's likely to be a high risk of harm to the individuals, in which case you need to tell the individuals as well. Now that is something you need to think really carefully about how you go about doing that. I find it much more difficult to draft employee communications sometimes than I do customer communications because employees, yes, they might not have the same level of grievance. They might not think, oh, right, this is an opportunity to get some money off my next coat or whatever. They don't think quite like that. Customers can possibly think, oh, data breach, let's see how I can use it to my benefit. Employees, not so much. But they definitely ask a lot more questions. They can be a lot more spooked, in my experience. Maybe it's different in your experience, Alex. So you have to think very carefully if you are going to be notifying employees, which in this particular situation, you probably will need to be notifying employees about how you do it.

That is why I always think it's quite important to get your comms team involved, your PR team involved. Also, employees may talk, they may go to the media, and then you might get, you know, you don't, you might get massively escalated, particularly if it's a no news day. So again, having your PR team to kind of control those external communications to the extent they get leaked is obviously really important.

I think just with employees, it's even more important than with consumers to have at least some kind of message that we're on it and we're solving it. And then if you are going to go down the credit monitoring or ID theft protection, you should have that in place and have a code ready. It won't remove the chance of a claim. As I said, there can be very small, rubbishy claims that can be brought. But just emotionally or employee relations-y, which is not a word, but you understand what I mean. It can really remove the chance of 100 complaints down to two.
If someone's like, oh, they're on it. Oh, sometimes we say we have reported already. Don't worry we're on it and if you're worried, here is a code at Experian, I'll just say. There are many other vendors available. Crowl, Western Union, whoever you want, whatever it is you know, 699, we're going to cover it. Here's the code for it. The number of times that really has just calmed people's nerves. And you then just get the usual two people who would have complained anyway. What are you going to do about them? And the beauty is with reporting to the actual data subjects, it just says without undue delay, as opposed to 72 hours. So you have a, that doesn't mean forever, but you have a much longer window to get your ducks in a row. So I think that is a very important point.

Yeah, absolutely. And then it's not just about the regulator and the data subjects. So your insurers, key person to notify. But then to the extent your business is regulated, so if it's a financial services business, then you might have the FCA. In our case, we would probably have to notify the SRA if it was a breach in the law firm. But if there are other regulators, you might need to notify as well. Then you might, depending how bad it is, think about notifying the police. Now obviously from an employment perspective we wouldn't recommend going straight to the police. There is a fair amount of process that you would need to put in place before you did that but that should also be potentially in the back of your mind, particularly in a case like this one.

I might just add one thing about the police point. A lot of clients are very keen to write to Max about how we're going to go to the police on the one hand, but sometimes we're discussing on the other hand, do you discuss with your hands? I don't know, I do, but we're sort of trying to downplay the breach and maybe we're not going to report it because we don't feel there's a risk. But there's a contradiction there that we're writing to one person, this is so serious, we're going to go to the police. But on the other side of the coin, we're not actually reporting. So it's strange, we do deal with that a lot.

I then start talking about things like, are we actually blackmailing Max if we're not really going to the police as well? So that is an absolutely nuclear solution, even though it did happen in the real case.

No, and to be honest, the point I was going to raise later, but I'll raise it now since you've made it, is that, I've just told you, yes, you should report to the because it's going to be a risk of harm. But there are lots of clients that often feel, do we really need to report something like this? Because they don't want it to show up some deficiencies within their own organisational systems. And I think that leads on to the next thing of, this is likely to result in some form of potentially regulatory action if you have reported it, potentially some form of litigation, if employee is feeling aggrieved. And then it's a question of who's responsible for this. Are you, as the data controller, responsible because did you have appropriate organisational and technical measures in place to prevent someone from doing something like this?
I would say no one has organisational technical measures in place to prevent someone from deciding to go off on a whim of their own and download loads of data. But there are certain measures that you could have seen happen. So for example, with data loss prevention software, you could see if someone's doing massive streams of downloads. And there's a case I'm going to talk about in a minute where this person wasn't downloading all in one go. He was downloading large quantities of data over a period of time. 

And those kinds of tools are very much, I would say standard practice now. So if you haven't got those in place, could you have prevented this somehow, or at least limited the damage somehow? So there is this potential that you as the data controller, even though this individual was not instructed by you to do this, could still have responsibilities under the Data Protection Act. 

But then on top of that, then there is this potential for vicarious liability. Now, Morrisons is a really great case on this particular topic, because it obviously goes through about whether or not the individual was acting within the field of his employment and whether what he was doing was something that was closely connected to the field of his employment. Now in this particular case, in the Morrisons case and probably in this case, you would argue that really no, they have gone off on a complete frolic of their own. They're clearly disgruntled. They're clearly doing something that they have not been in any way, shape or form asked to do. And so, you know, hopefully in this scenario, you'd be able to point to Morrisons and say, you know, it's not us. We shouldn't be responsible. But unfortunately, vicarious liability is not an easy topic to always just say there's a black and white answer, there's always an if and a but, it always could depend on this and depend on that, so it's always a threat that you need to think about. 
And then on top of that, the individual themselves, the fact that they've done this, this will no doubt attract some form of criminal liability as well. So whether it be under the Theft Act, under the Data Protection Act, there's a specific provision around unlawful acquisition of data, whether the misuse of private information, whether under fraud. So there's various different angles you could go at the individual from a criminal liability as well. And then of course, there was some kind of precarious liability, then you as an organisation could have that criminal. So it does can get quite complicated for you quite quickly.

Just one point to talk about, I'm just interested in people's thoughts as well. it's the width of his access. It is dangerous for anyone to have such wide access. But I do appreciate that occasionally it needs to happen. But if you're ever involved in custodian limiters, it really should be people should really only have segregated access. The second point is, of course, what we really should have done here was, I don't know, reduce his access, because it was quite a serious offense. But who does that? I know it happens during redundancy consultations. People talk about it. And you always have the difficult discussion about, well, should you go home and you don't have any system access? But then here, if you were going to do that for every single person who might be disgruntled, probably no one would ever have any access in any organisation. So it is a very difficult thing. But the final point I'll make on that, we're talking about insider threat here. But this kind of enormously wide access also has an outsider threat issue.

Many of the ransomware attacks or similar arise because someone was socially engineered to give their passport and that person had access to so much, it allowed, for want of a better word, the hackers to come in and access everything. Whereas of course, if they're only segregated, it would have only been a limited exposure for the business. So you can see it all fits together. This is insider a threat and it's caused this issue. But the number of times we see outside a threat where, yeah, they just, we're very surprised that getting in here allowed them access to everything, even in the biggest organisations.

Yeah. And then because of the fact that there is potential litigation, regulatory action here, the other thing to be really mindful of is when you're having your internal discussions about dealing with this is privilege and maintaining privilege. Because that will be absolutely instrumental further down the line if this incident does end up having the long tail that we hope it doesn't have, but could potentially have.

I might do just one minute on that because it's quite an important point. Investigations are happening in the heat of the moment and you're having to do something open because you might need to send it to the ICO. You might be challenged on it by an insurer, who knows? But of course, you might engage a third party security consultant who then writes something that later on is being used to found data claims in litigation.

So you can't do something in the ether, it can't all be privileged. Because you don't have anything to say you’re doing. But the only way you can really solve that is to have a set protocol already and people who've already understood this or done tabletop exercises, Bryony, you're not thinking about it live in a real situation. You've actually run through, I know people hate tabletop exercises and they're a bit weird, aren't they, but I mean, the idea of this is so you've already thought about these issues before and everyone knows what their role is. You know, even bring your third party security consultant in so that they understand.
Let's talk orally about what your findings are going to be before you write on a piece of paper that is there in perpetuity “this is the worst breach I've ever seen. You're all gone.” Which obviously no one would write. But it is just a very important point.

So I guess then thinking once you've won privilege, what other things do you need to think about? And I think here it is that moment of, mean, first of all, mitigate, mitigate, mitigate. But then it's that moment of kind of reflection. Once you have notified the ICO, the thing that they'll be interested in, the thing that will help you in the event of a claim is what you've done afterwards to stop this from happening again. And so that is when you then kind of relook at kind of the facts. You look at things like access, at things like data loss prevention tools. You look at what disciplinary protocols you have in place that this person felt that maybe in this particular case this person didn't really care. And therefore, no matter what disciplinary protocols you have in place, they were going to ignore them, but sometimes the fact that there's something a bit more meaty on the bones might make them go, actually, we really don't want to do this. You might want to look at also what contractual levers that you might want to pull within your employment contracts.

But just making sure that these sorts of things don't happen again. As I say, it is very difficult to prevent. But if you are going to be going to the ICO about this, which most of the time I would say you would need to in this kind of scenario, you want to be presenting a picture that you are a good data protection citizen, and this person has very much decided to go off on their own. And then as I say, you just need to manage that comms with your workforce to make it really clear this was a one-off event, but we're going to protect you going forwards.

And although we've mentioned the ICO, most of our businesses are international, so the payroll could have French, Middle Eastern, Singapore. There's all different ways you could do that. Are you going to report to 27 different jurisdictions? Probably not. Are you going to report to the one headquarter jurisdiction? Are you going to go around the world taking advice? It is difficult, and these choices have to be made very quickly. In Europe, are you going to report to all 27, or are you just going to report to one now, I think, is where we're heading.

It is a bit of a nightmare. One final point on the dark web I just want to make. These things never end. You're not cleaning the dark web. The data is there. There will probably be white hack hackers will come to you in a year, five years, six years, say, do you know your data's here? We have it all the time. I've just noticed this. Will you pay me a bounty? So I'm afraid it's the gift that keeps on giving. Because at some point, you just have to say, we have done everything that we possibly can. And then you just, you are satisfied because otherwise it's never ending.

Yeah. The one case that I referred to earlier, which isn't on the dark web, but it's a live case at the moment with British Gas, where one of their employees, who was I think one of their call centre employees, was downloading loads of data of, I think it was customer data, but then selling it to marketing companies, I think for 45,000 pounds or so, so not a huge amount, obviously for that person it was, and they're actually facing criminal sanctions at the moment, and it will be really interesting to see. We haven't heard anything about whether or not British Gas have been investigated or whether there are claims against British Gas. But that, again, is another very interesting one and a very live example that's hit.

And this, Morrisons guy went to prison for two or three years. I mean, he had a custodial sentence. But maybe there needs to be more to promote that by the police that, you could go to, you can go to prison for this. It is a very serious thing.

So I guess the last thing that we kind of just wanted to talk about, move on to the flicker, is the AI literacy learning, one of the things that came out of the, annoyingly, we would have done the case studies the other way around, but because of the scenario and the sequence of events, we had to do it this way around. So we're of jumping back to AI for a second. 

One of the key things around this sort of use of shadow AI and preventing use of shadow AI is ensuring that you have AI literacy within your workforce. Now, if you've been eagle-eyed wandering around here today, you will have seen that we actually have just launched our AI literacy e-learning module which is aimed at specifically helping the whole workforce, so it's not just aimed at legal teams, understand the risks, the rewards of AI, but really the risks of AI, how AI works, and then kind of the limitations of it, that then they can then really understand the importance of good governance and why it is so critical for having to comply with policies and things like that, so scenarios like our first scenario don't happen.
It is actually also a legal requirement to the extent that you are operating in the EU. It is now a legal requirement to have AI literacy training within your workforce. So it was just something that we wanted to flag. And I have to say, there are other tools on the market as well. But certainly, in terms of that first case study, that is a really, really live threat at the moment and AI literacy is a really good way to kind of make sure that people are taking note of some of the risks of using AI tools and actually reading those policies that you send around. And as I say, my big tip there is do's and don'ts, short and sweet, nothing more than a couple of pages that people can just go, OK, right, I shouldn't be doing that, or should be doing this. And that will really help mitigate the first scenario.

That was one of the live sessions recorded at the Work Agenda 2026. We hope you enjoyed listening as much as we did delivering it on the day.

Don't forget you can visit our website, lewissilkin.com to find out the latest on the issues covered in this episode, plus those from across the series. And if you did enjoy it, please share with your colleagues and friends, like and add any comments you have. We'd love to hear them. For now though, thanks for listening and goodbye.