AI In Corporate Audits Is Real, So Are Hallucinations

There is a flip side to it, though. Machines can also confidently invent facts, fabricate citations, and misinterpret data: a phenomenon known as AI hallucination. 

Hallucinations are AI’s Achilles' heel, and even the Big 4 were caught in the mesh. In 2025, Deloitte agreed to partially refund (about $290,000) the Australian government after a report it generated was flagged for alleged AI hallucination. 

It didn’t stop with Deloitte. In 2026, EY Canada withdrew a 44-page cybersecurity report, Points of Attack: Uncovering Cyber Threats and Fraud in Loyalty Systems, after it was pulled up for reportedly fabricated data generated by AI.

These episodes serve as a reminder that while AI can accelerate analysis, it cannot replace human judgement or accountability. Amid rising instances of AI hallucinations and inaccuracy, the Institute of Chartered Accountants of India (ICAI) has been taking concerted measures to implement stringent regulations for the responsible use of AI in audits. The ICAI stresses that AI must augment human workflow while human intervention is necessary for final decision-making.

This decision is directly applicable in corporate audits. Companies are increasingly deploying specialised AI tools to analyse entire datasets in real time, detect anomalies, and automate compliance workflows. 

Yet auditors argue that despite AI's speed and scale, compliance cannot be left entirely to machines. The final responsibility for interpreting findings, exercising professional judgement, and signing off on audit opinions still rests with humans.

AI fetches data, identifies patterns, and produces reports. It delivers instantly. For companies, this is a dealbreaker, as they no longer have to rely on periodic, either quarterly or annual, audits to determine anomalies before they make a lasting financial impact.  

AI in audit has enabled real-time flagging of anomalies.

But AI cannot reason. It cannot be held accountable, either. Most of all, it cannot be blamed for the anomalies the machine itself introduces. 

As more and more companies harness specialised AI tools for audits in real time, compliance auditors stress the relevance of human intervention. 

Instead of manual testing or physical sampling, companies have been utilising “Exceptional Exceptions”, which is a combination of rules-based and machine learning tests that allow users to run analytics on general ledger data, for audits. 

The transition is well evident: companies have increasingly been adopting AI in their processes, particularly in audit. 

David Marquis, Chief Executive Officer, Caseware had remarked: "With two-thirds of firms already embedding AI into their strategies, the profession has crossed a pivotal threshold. The question is no longer whether to adopt but how to deploy AI in a way that the profession can truly depend on." 

Even the Big Four no longer rely entirely on manual, sample-based auditing, as they have adopted data-driven audit models. According to experts, major banks, insurance corporations, manufacturing firms, and public sector organisations throughout the world are also leveraging advanced audit analytics, automation, and AI-assisted audit solutions to improve risk management and compliance.

The rate of AI adoption in audits is highly skewed in favour of the early adopters - the Big 4 or Big 7 firms.

Earlier, corporates relied on traditional and annual audits to evaluate risk. AI and specialised tools have upended it. As risks evolve, particularly cybercrime threats, organisations have been scrambling to seek specialised solutions to flag anomalies in real-time. Platforms that offer full-population data analysis, automated audit workflows, and stronger audit documentation have come to the forefront. 

According to a recent report, AI in Audit Market, the global AI in the audit market size is expected to be worth around US$11.7 billion by 2033, from US$1.0 billion in 2023, growing at a CAGR of 27.9% during the forecast period from 2024 to 2033.

Audits have progressed from manual, sample-based examinations to data-driven, ongoing analysis. Auditors can use specialised audit technologies and artificial intelligence to analyse whole datasets, find abnormalities, automate repetitive activities, and provide deeper insights, thereby enhancing audit quality and efficiency.

"AI in audit would likely make strides in terms of quality assessment. This can mostly be internal. But in the case of a compliance audit, human intervention is still a prerequisite, as AI cannot, obviously, be held accountable for oversight errors. AI could be harnessed for compliance, but it has to be monitored thoroughly," says an auditor associated with Deloitte. "Both annual and quarterly audits are still made for compliance reports, but the increased use of AI in quality and internal assessment is visible," he further says. 

According to a report, Benchmarking Generative AI In Internal Audit, 42% of Chief Audit Executives (CAEs) see embedding GenAI into the audit department’s workflow and methodology as an important priority for 2025. 

AI usage in auditing is constantly increasing," Deepjee Singhal, Co-Founder, Sama Audit Systems & Softwares Pvt. Ltd., told The Secretariat.

According to industry assessments, the vast majority of audit and accounting companies are either actively employing or investing in AI-enabled solutions for data analysis, risk assessment, document review, and workflow automation. Adoption is especially high among multinational corporations and professional services firms.

Some of the specialised AI-based tools are: CaseWare IDEA (for data analysis and audit testing); CaseWare Cloud (for audit management and collaboration); ACL Analytics (provides data analysis and ongoing auditing); Microsoft Power BI (for visualisation and reporting); and Alteryx (for data preparation and automation).

"The majority of these tools do not rely entirely on artificial intelligence. They mix traditional analytics, automation, and rule-based testing, with AI elements being added to help with risk detection, anomaly identification, and predictive analysis," Singhal pointed out.