The Modern Investigations Roundtable: Key Takeaways

As AI increasingly becomes both a source of risk and a powerful investigative tool, organisations are navigating new challenges relating to fraud, whistleblower reports, ethical considerations, and information governance.

On 14 October 2026, Ethical eDiscovery and Punter Southall Law (PSL) brought a select group of senior professionals together for a collaborative roundtable discussion on how organisations are adapting to this rapidly evolving landscape.

In the room was a mix of professionals, including, among others, in-house counsel, heads of compliance and barristers. Jonathan Armstrong of PSL and and James MacGregor of Ethical chaired the discussion.

Setting the stage was simple, as everyone in the room could agree that investigations and subject access requests (SARs) are growing in volume and cost. The question was whether a careful use of technology is part of the answer. The topic quickly evolved into a lively discussion. Below are the themes that came up most.


About this article. The roundtable was held under the Chatham House Rule, so comments are not attributed to individuals or their organisations. What follows reflects views and experiences shared in an open discussion among professionals. These are not necessarily the views of any attendee's employer, Ethical eDiscovery or Punter Southall Law. This is not legal advice.


1. Using technology to speed up investigations

Jonathan Armstrong and James MacGregor opened the conversation with some of the cases they had worked on together, including a fraud investigation spanning several jurisdictions, among them the Middle East, where early results were needed fast. The technology they used helped them identify likely conspirators sooner and bring interviews forward, reaching a quicker and cheaper answer to the questions that mattered.

Jonathan outlined several core advantages of using technology in investigations:

  • reducing cost and speeding up lower-value "grunt work"

  • proximity analysis, for example working out who knows whom

  • streamlining routine tasks, for example identifying who was working outside their contracted hours

  • using AI to assist review in areas such as subject access requests

  • emotional analysis

  • threading across media: when individuals say in an email that they are moving a discussion to WhatsApp or Slack, technology makes it easier to see those messages in chronological order rather than in separate pockets

2. The rise of AI-assisted subject access requests

A recent survey found that 86% of European organisations had received a SAR drafted with the help of generative AI, and the average cost of handling SARs for medium to large companies is now reported to be €680,000 per year.

This matched what participants were seeing. One organisation had recently received a SAR that was 67 pages long and drafted using generative AI. Some participants also agreed that AI can fuel a spiral of requests and embolden the people making them.

One of the strategies discussed for handling this SAR work was using AI in a way similar to technology-assisted review (TAR), which helps to reduce that cost.

3. Guardrails, quality and defensibility

At the same time, guardrails remain essential. Jonathan referred to the Solicitors Regulation Authority's (SRA) recent warning notice on the misuse of AI, particularly around the use of in-house tools. The discussion covered the need to:

  • ensure quality and follow a proper methodology

  • consider privilege

  • make sure the tools used are appropriate, for example by not relying on free licences

  • be cautious where an AI tool has been commissioned by the business rather than by lawyers

Sampling is an important way to show that the AI is doing its job. A sample of 10% seems to be a potential emerging best practice, with PSL currently checking 15% of non-relevant documents in SARs, partly because of how unpredictable regulators can be.

Participants also discussed the value of a written methodology and defensibility statement when responding to SARs, so that an explanation is ready if the Information Commission's Office (ICO) or another regulator asks for one. When asked whether AI was used to draft the prompts that categorise data as relevant or not relevant, it was confirmed that the prompts are drafted manually.

4. Privilege and the use of AI in litigation

An example shared was a query that one barrister had received just a week before the roundtable, asking for advice on the use of AI and the loss of legal professional privilege. Another point raised was that the risk to privilege is being used tactically by aggressive litigators, for example by seeking disclosure of the prompts used when AI is involved, with some in the room agreeing that AI is being weaponised.

Another interesting highlight was the lack of proper infrastructure for AI in the legal system, which one barrister compared to when the Proceeds of Crime Act (POCA) came into force and the judiciary had not been trained to deal with it. AI-generated instructions to counsel are also on the rise, which raises privilege concerns and is becoming a point of contention in various pieces of litigation. Many people, including some lawyers, might not yet understand prompts or the risks of prompts in disclosure. On the investigation side, there are also concerns about privilege where non-legal personnel are involved.

AI notetakers were a particular concern, especially as privilege differs between jurisdictions. One in-house team is advising colleagues to avoid them during privileged conversations, even though some team members are writing their emails using AI. Jonathan shared recent cases at PSL that showed the considerable downsides of AI notetaking, and described it as another example of organisations holding data without proper prior planning.

5. In-house perspectives on adopting AI

Views on AI varied widely among the in-house participants. Some are proceeding with caution. One business has been under pressure to adopt a company-wide AI solution (Claude) but has so far resisted. Another has not yet used AI in any compliance investigation and is still weighing up the right level of involvement, and whether it risks missing out. Others are further ahead, with AI already used for regulatory compliance, and an expectation that it will play a bigger role in the public sector and education.

Where AI has been adopted, the experience has been mixed. One organisation has built AI agents based on Claude to help with SARs, deciding where to use them based on how aggressive the regulator is in each jurisdiction. Crunching data is faster, but the overall workload has increased three to fourfold, as the agents need ongoing training and management. Accuracy and the risk of circular endorsement remain concerns, raising the question of how regulators can keep up.

Choosing the right provider is a challenge in itself. Participants felt the market for AI start-ups is oversaturated, which makes it hard to screen providers, plan for one going offline, and find reliability and transparency. Air gapping is also increasing, and data sovereignty is becoming a key issue.

6. Data volumes, confirmation bias and the quality of investigations

The volume of data now involved in investigations prompted one of the most reflective parts of the conversation. There was concern that, under pressure to reach a conclusion, confirmation bias is becoming a real issue, and that data can end up telling people what they want to hear rather than what is really there. As one participant put it, borrowing a well-known joke, a statistician uses data like a drunk uses a lamppost: for support rather than illumination.

Jonathan agreed that with so much data, it is possible to find some support for almost any theory. He described a past case in which he responded to a whistleblower's report to the Serious Fraud Office (SFO) by presenting further data from the investigation. This showed where the whistleblower's data was inaccurate, or at least put it in context.

There was a general sense that the quality of investigations has dropped, even as the work involved has grown. Before getting to the core of a matter, advisers often first have to establish what the client is trying to achieve and what material they actually have. Participants also worried about the effect on skills: if the next generation never does the work manually, they may never reach the same standard.

Data retention was another theme. Spain's whistleblowing rules, under which an organisation should keep information only for as long as it is needed to decide whether there is a case, were seen as a good approach, in contrast to a UK tendency to hoard data. Large volumes of data also create problems with confidentiality and need-to-know, as everyone feels they should be included because some of their data is being used.

7. How regulators are responding

There was interest in how the ICO is coping with the rise in SARs, and Jonathan shared PSL's insight into the issues facing the ICO and other regulators in the GDPR zone. The ICO may be looking at AI-assisted triage, and one German regulator already has a similar pilot project that it is looking to share. Certain issues can prompt the ICO into action more quickly, as it has priority areas such as children's data or the commission of criminal offences.

Concerns were also raised that financial regulators are leaning into the use of AI. Those regulators have rules against disclosing confidential information about the firms they regulate, yet that information is being fed into very large corporations. Over time, those corporations could gather enough pieces to build a whole picture, which raises questions about national and public security.

Italy's new AI law also came up. It includes provisions on protecting professional dignity.

8. Law firm fees

Some participants are seeing law firm prices increase, and clients are asking how firms are using AI and whether this will lead to a reduction in fees.

Where the discussion landed

Participants agreed that there is no such thing as a perfect investigation.

Ethical eDiscovery and Punter Southall Law thank everyone who took part and spoke so openly. To hear about the next roundtable, or if you have any further questions or would like to add to the discussion, please contact Sabrina Araujo at sabrina@ethicalediscovery.com.

Further reading

Next
Next

Data Subject Access Requests in the age of AI: Challenge, opportunity and practical considerations