Balancing innovation & risk

Ruby Hamid of Ashurst examines the use of AI in Investigations. Building on decades of e-discovery experience, AI is providing new tools and techniques to save time and reduce costs.

The rise of generative AI has prompted questions about how AI can be used to improve, expedite and reduce the cost of investigations. The Ashurst Global Investigations team has a busy caseload across the Middle East and is working hard with Ashurst’s digital experts to identify solutions.

Investigations in Middle East corporates are on the rise. Regulators are increasing pressure on companies through expanded regulatory regimes. Think the new AML regulations which came into force in Dubai in mid-October, the recent Personal Data Protection Law in KSA, and the Whistleblower Protection Regulations 2024 in the ADGM. Expanded regulatory expectations mean regulator-led investigations when things go wrong. Combine that with increasing compliance and governance expectations from employees, shareholders and foreign JV partners, and you have corporates conducting internal investigations more than ever before.

AI IN E-DISCOVERY

Investigations mean data, and the use of technology to interrogate it: e-discovery. Fortunately, e-discovery has been a core part of the conduct of investigations for decades, and it is one of the areas where Large Language Models (LLM), a category of Generative AI, are bolting on to existing Technology Assisted Review tools (TAR) in new and exciting ways.

Imagine this scenario: a whistleblower reports comes to you, as GC of a large company, alleging that a sales team has been pushing huge contracts to a particular supplier, in breach of the procurement policy, because the owner of the supplier is in a relationship with the sales director. Kickbacks are rumoured to be significant. The whistleblower saw an email which concerned them, and knows the pair have been messaging on Teams during working hours. What would have been a standard e-discovery exercise (pull email and Teams data, look at financial records, identify examples of unusual or troubling communications) can now make use of a LLM. A review of relevance can be sped up, pushing the communications most likely to be relevant to the eyes of reviewers, more quickly. An LLM can code documents for issues or categories, enabling investigators to focus on the data likely to matter most. Personal data can be stripped out more effectively than using the old-fashioned key-word search, protecting your company from contravening data protection legislation.

USE OF LLMs ACROSS INVESTIGATIONS

LLM can be used in other ways. The investigation team can devise natural language queries to identify key events, custodians, topics, and build chronologies of events. Those chronologies can help interviewers get to key facts more quickly with witnesses, and might speed up the preparation of disclosures to enforcement authorities if, for example, there is a money laundering risk, or a suspected breach of financial sanctions.

AI can help with the set up and scoping of an investigation by summarising internal policies (for example, a quick scan of the whistleblowing policy to enable lawyers to advise on setting up a policy-compliant investigation), and by creating schedules to track timelines and evidence gathering. New tools can transcribe interviews without creating recordings, important as many investigations specialists advise against creating recordings of witness evidence, to protect against future discovery and disclosure risk. A machine transcription will not be quite right – names are not picked out and the text can be messy. But there it is a huge timesaver to have an approximation of an interview note, instantly, for lawyers to adjust in much less time.

Translation software is of particular benefit in cross-border investigations, which the practice in the Middle East often involves. Imagine conducting an investigation into alleged bribery in your company’s Indonesian subsidiary without having to send every communication to a third-party translation vendor, before you can start interviewing your key witnesses.

A further area is the creation of summaries for key internal stakeholders. The headline points of an interview or a document produced instantly, so that you can add the value of your legal experience by bringing in the nuance, or the advocacy – as the situation requires.

PROCEEDING WITH CAUTION

Two major concerns we are seeing are confidentiality and accuracy.

Confidentiality is an area which many companies are – rightly – interrogating. Will their data remain secure if an LLM is applied to it? Will it be ringfenced, so that it is not passed back into the LLM for it to continue to train itself?

Accuracy is a prevailing challenge where open AI is used, scraping the internet for information without being able to weigh up the credibility of the content, the reliability of the author, or the veracity of the source. This is an area of particular concern for lawyers, as legal regulators around the world bring disciplinary proceedings against lawyers who have relied on hallucinated caselaw, purportedly identified as relevant by AI.

A solution is to direct AI only at the data universe of the investigation. For example, AI which searches solely the documents in your case to create a chronology (rather than bringing in untested information from the web), or summarises only your company’s procurement policy (rather than all procurement policies which can be found online).

The headline is that AI can – and should – form part of an investigation, but cannot replace trained and experienced investigators, or legal advisers. Efficiencies can be made in investigations, from scoping and data review, through evidential analysis and reporting.

And the future? Lawyers and legal technologists working hand-in-hand, training models to provide reasons for identifying data as relevant (which can be interrogated and considered), developing workflows which strip administrative tasks from investigators, leaving them able to focus on the essential practice of finding facts, understanding risk and – ultimately – improving corporate culture and performance.

This article was written by Ruby Hamid, and first published in the magazine, the Oath, the Middle East law journal for corporates in November 2025.