Cifas protection against fraud

Fraud often crosses organisational boundaries. A criminal might use a stolen identity to apply for credit, open a mobile phone account, and make an insurance claim, each with a different company. Viewed separately, each attempt may be difficult to recognise.

Cifas, originally known as the Credit Industry Fraud Avoidance System (CIFAS), helps participating organisations share fraud risk information so they can identify those connections and investigate before further harm is done.

What is Cifas?

Cifas is a UK not-for-profit fraud prevention organisation. Its members include organisations in sectors such as banking, lending, insurance, telecommunications, and the public sector. Members use Cifas services to share information about identified fraud risks and to check applications or existing relationships against information contributed by other members. Cifas also offers protection for individuals whose personal details may be misused.

The best-known service is the National Fraud Database. It contains records covering risks such as identity fraud, account takeover, false applications, false insurance claims, and misuse of an account or facility. Cifas also operates a separate Insider Threat Database to help members address serious fraudulent or dishonest conduct involving staff and prospective staff. These services have different purposes and filing rules.

How does the process work?

A member organisation may search the National Fraud Database when assessing an application, reviewing an account, or investigating suspicious activity. If a relevant record is found, it gives the organisation information to examine. The organisation must still establish what the information means for the particular case and make its own decision.

Members also contribute cases. Cifas’s National Fraud Database rules require reasonable grounds to believe that fraud or financial crime has been committed or attempted. The evidence must be clear, relevant, and rigorous, and the conduct must fit a defined case type. Further filing conditions apply, including conditions concerning the product or service involved. A suspicion without adequate evidence should therefore not be treated as sufficient grounds for a fraud risk filing.

A Cifas record is sometimes called a marker. That informal term can be misleading because the type of record matters. A record concerning suspected fraudulent conduct has a different meaning from a warning that someone is an innocent victim of impersonation or has registered for extra protection. Treating every Cifas match as evidence that the named person committed fraud would undermine the very protection the system is intended to provide.

How does Cifas protect businesses?

Information sharing helps a business see fraud risks that may already have appeared elsewhere. For example, if a criminal uses stolen personal details to apply for several products, a member assessing a later application may be alerted to an earlier identity fraud case. The business can then carry out additional checks before granting access to credit, goods, or services. Members can also use the database to identify risks involving account takeover, false claims or misuse of accounts.

This can reduce direct financial losses and the time spent resolving fraudulent activity. It can also protect genuine customers: identifying an impersonation attempt early may prevent an account being opened in their name. The Insider Threat Database serves a related business purpose by allowing members to assess qualifying risks involving staff or prospective staff, subject to its own evidence and filing requirements.

The value of the system depends on careful use. Businesses need sound investigations, accurate records, appropriately trained staff, and a proportionate response to any match. Cifas states that National Fraud Database records must be supported by accurate, relevant evidence and that access is restricted to trained staff.

How does Cifas protect individuals from identity theft?

Identity fraud occurs when someone uses another person’s details to obtain a product or service or gain access to an existing account. A victim might discover an unfamiliar application, receive correspondence for an account they never opened, or notice activity they do not recognise. By sharing information about such attempts, Cifas members can help other members spot further misuse of the same identity. Members can also record that the genuine person was a victim of impersonation, making clear that they are the innocent party.

For someone at heightened risk, Protective Registration provides another route. It places a protective flag against the person’s details so that participating organisations can take extra care when those details are used. This may mean additional identity checks and, for a genuine application, a longer process. As of September 2026, Cifas lists the service at £30 for two years. Cifas says Protective Registration does not affect a person’s credit score. It is not credit monitoring, insurance, or a guarantee that every attempted fraud will be stopped.

Someone who believes their identity has been used fraudulently should also contact the organisations involved to secure affected accounts and correct their records. Cifas advises victims to report identity fraud to the appropriate fraud reporting service. Protective Registration adds a warning for participating organisations; it does not replace those immediate steps.

What if a Cifas record is wrong?

Because a fraud risk record can have serious consequences, people need a way to see and challenge information held about them. An individual can make a free data subject access request to Cifas for a copy of information held about them in its fraud risk databases. Cifas says it must respond within one calendar month of receiving a fully completed request and supporting documentation.

If someone disputes a record, Cifas directs them first to the organisation that filed it and to that organisation’s complaints process. That organisation can remove the record if it upholds the complaint. If the matter remains unresolved, the person can obtain the organisation’s final response and ask Cifas to review the information, provided they have also completed a data subject access request.

Shared information, careful decisions

Cifas works by helping members act on fraud risk information beyond their own records. For businesses, that can reveal repeated attempts and support earlier intervention. For individuals, it can help stop stolen details being used elsewhere and provide a visible warning when extra identity checks are needed.

Its protection depends on two equally important steps: members must file well-evidenced, accurate information, and they must investigate what a match means before deciding how to respond. Used carefully, the process helps prevent fraud while protecting the people whose identities fraudsters try to exploit.

Identity and Access Management for AI Agents

AI agents are changing the identity security landscape. Unlike conventional applications, agents can interpret objectives, access sensitive information, invoke tools, and take action with a degree of autonomy. They may work across multiple systems, collaborate with other agents, and alter their behaviour as circumstances change. This creates significant opportunities for enterprises, but also introduces a new kind of identity risk.

When considering the challenges of managing identities and permissions for AI agents, I found myself reflecting on complex IAM projects I had undertaken in the past. One recurring challenge was establishing which user accounts belonged to which individuals across numerous applications, often with inconsistent naming conventions and no straightforward means of correlating accounts to a single identity. These experiences reinforced the importance of establishing identity, ownership, and accountability before attempting to manage access effectively. Although AI agents introduce new complexities, many of the underlying IAM challenges are familiar.

Many Identity and Access Management (IAM) systems were designed for a more predictable environment. A person signs in, an application requests access, or a service account performs a predefined function. Permissions are often assigned in advance and remain in place until somebody reviews or revokes them.

AI agents challenge these assumptions. Their activity can be dynamic, their access requirements can change from one task to the next, and their actions may occur at machine speed. Establishing trust at the beginning of a session is no longer enough. Organisations need to understand whether that trust remains justified throughout the agent’s activity.

As AI agents become part of everyday enterprise operations, identity security must evolve in three important ways. Organisations must:

  • Continuously understand identity risk
  • Reduce standing privilege through context-aware access
  • Establish trusted identities for AI agents

AI agents are joining the enterprise identity estate

An AI agent is not simply another piece of software running in the background. An agent may be able to retrieve customer records, analyse internal documents, update business systems, write code, initiate workflows, or communicate with employees and customers. More advanced agents may plan sequences of actions and select the tools needed to complete an objective. Those capabilities make an agent an active participant in the enterprise. From a security perspective, it must be treated as an identity.

That identity must be discoverable, authenticated, and governed. The organisation needs to know who created the agent, who owns it, what purpose it serves, which systems it can access, and what actions it is authorised to perform. It must also be possible to determine which human, application, or agent initiated a particular action. Without this foundation, organisations risk creating a growing population of powerful but poorly understood identities.

This is not an entirely new problem. Enterprises already manage human users alongside service accounts, workloads, machines, application identities, and other non-human identities. Many organisations continue to struggle with excessive permissions, unclear ownership, and credentials that remain active long after their original purpose has disappeared.

AI agents add another layer of complexity. They can act more independently than conventional non-human identities while operating with less direct oversight than human users. They therefore need governance that reflects both their technical nature and their potential autonomy.

Traditional access models are reaching their limits

Many access models are based on a relatively static idea of trust. An identity is authenticated, assigned a role, and given access to a defined set of resources. Periodic certification processes are then used to determine whether those permissions remain appropriate. This model already creates risk when people change roles or service accounts accumulate access over time. In an agentic environment, its limitations become more pronounced.

There is also an important parallel between human users and AI agents. A human employee may have legitimate access to a system but use that access beyond the responsibilities or authority associated with their role. An AI agent presents a similar challenge. Technical capability to perform an action does not mean that the action falls within its authorised mandate. The difference is that an autonomous agent may act at greater speed, across multiple systems, and without the same opportunities for human intervention. The fundamental IAM principle remains unchanged. Access must be governed by what an identity is authorised to do, not simply what it is technically able to do.

An agent’s legitimate access requirements may vary considerably between tasks. A customer-service agent might need to read a customer record to answer a question, update an address following explicit approval, and initiate a refund only when additional conditions have been satisfied. Giving that agent broad, persistent access because it might occasionally need those capabilities increases the potential impact of an error, manipulation, or compromise.

The issue is not necessarily whether the agent should ever be allowed to perform a sensitive action. It is whether that permission should exist continuously. Identity security must therefore move beyond asking, “Does this identity have access?” It must also ask:

  • Why is access being requested?
  • What objective is the agent pursuing?
  • Who or what initiated the task?
  • Which data and tools are involved?
  • Is the requested action consistent with the agent’s intended purpose?
  • Has the identity’s risk changed since access was granted?
  • Should additional approval or verification be required?

These questions cannot be answered effectively through authentication alone. They require an ongoing understanding of identity, entitlement, behaviour, and context.

Continuously understand identity risk

Organisations need a unified view of risk across human, non-human, and AI identities. That starts with visibility. Security teams must be able to discover the identities operating throughout the enterprise and understand the relationships between them. This includes the permissions they hold, the credentials they use, the resources they can reach, and the people or systems accountable for them.

For AI agents, visibility should extend to their purpose, tools, data sources, and operating boundaries. An organisation should be able to distinguish between an approved production agent, an experimental agent created by a development team, and an unregistered agent connecting to corporate systems. Discovery, however, is only the first step.

Identity risk changes continuously. An agent may begin a task under normal conditions and later request an unusual permission, access a sensitive repository, or invoke a tool it has never used before. Its credentials could be exposed, its instructions could be manipulated, or its behaviour could diverge from its established purpose. Risk assessment must therefore be continuous rather than periodic.

Signals can include the sensitivity of the requested resource, the identity’s current privileges, the origin of the request, recent behaviour, unusual tool use, and the relationship between the requested action and the assigned task. The objective is to build a current picture of risk and use it to inform access decisions.

This does not mean treating every deviation as malicious. Agents may legitimately encounter new conditions as they work. The security model must be able to distinguish reasonable adaptation from unacceptable risk, and respond proportionately.

Depending on the circumstances, that response could involve granting access, limiting functionality, requesting approval, requiring stronger verification, or stopping the activity altogether.

Reduce standing privileges

Permanently assigned privilege is particularly dangerous in an agentic environment. When an identity permanently holds more access than it needs, every error or compromise has a larger potential impact. With AI agents operating quickly and across connected systems, excessive privilege can allow an isolated problem to spread before a human has time to intervene.

Organisations should minimise the permissions held by default and provide additional access only when it is required for an authorised task. This is the principle of just-in-time, least-privilege access: the right identity receives the right level of access to the right resource for the minimum necessary period.

For an AI agent, that could mean receiving permission to read a particular dataset for one task rather than having permanent access to the entire data platform. It could mean being allowed to prepare a transaction but not approve it, or to invoke a tool only after a human has authorised a sensitive step. Crucially, these decisions should be based on real-time context rather than static role membership alone.

The same request may present different levels of risk depending on the data involved, the agent’s recent activity, the task it is performing, and the way the request was initiated. Context allows organisations to make more precise decisions instead of choosing between unrestricted access and complete denial.

Access should also expire automatically when the task is complete. This prevents temporary permissions from becoming permanent entitlements and reduces the burden on manual review processes. The result is a more adaptive model of privilege, one that supports useful automation without giving agents broad authority simply because they may need it in the future.

Establish trusted identities for AI agents

Every enterprise AI agent should have its own verifiable identity. Shared credentials and generic service accounts make it difficult to establish accountability. If several agents use the same identity, security teams cannot reliably determine which agent performed an action, who authorised it, or whether the activity was appropriate. A distinct agent identity creates the basis for effective governance.

That identity should be linked to an owner, a defined business purpose, and an approved set of capabilities. Its lifecycle should be managed from creation through modification to retirement. Credentials should be protected and rotated, permissions should be reviewed, and unused identities should be disabled promptly.

Trust should also extend beyond the identity itself. Before an agent is allowed to act, the organisation may need to verify the integrity of its operating environment, the tools it is invoking, and the source of its instructions. These checks help establish a chain of trust connecting the agent, the person or system directing it, the resources it accesses, and the actions it performs.

Policy can then be applied consistently. An agent can be granted access when its identity is verified, its purpose is authorised, the requested action is appropriate, and the current risk is acceptable. When those conditions change, its access can change with them. This makes trust conditional rather than permanent.

Preserve accountability and human control

Autonomy does not remove the need for accountability. Organisations must be able to reconstruct what an agent did, which information it accessed, which tools it invoked, and why a decision was made. Audit records should connect activity to the agent’s identity, its owner, and the person or process that initiated the task.

Higher-impact actions may also require meaningful human oversight. The objective is not to place a human approval step in front of every agent action. That would undermine much of the value of automation. Instead, organisations should define clear boundaries based on risk.

Routine, low-risk actions may be completed automatically within established policies. Sensitive or irreversible actions may require approval, segregation of duties, or additional verification. Activity that falls outside the agent’s defined purpose may need to be blocked and investigated. These controls allow autonomy to increase without allowing accountability to disappear.

Identity security becomes central to securing enterprise AI

AI agents will increasingly operate across applications, data repositories, and business processes. Trying to secure each interaction in isolation will become difficult as the number of agents and integrations grows.

Identity provides a common control point. By establishing which agent is acting, who is responsible for it, what it is trying to accomplish, which permissions it holds, and the risk it currently presents, organisations can govern agent activity consistently across different systems.

It connects visibility, authentication, authorisation, privileged access management, monitoring, and governance. It also gives organisations a consistent way to apply policy as agents move between tools and environments.

The purpose is not to restrict AI adoption. It is to create the conditions in which organisations can use agents confidently and at scale.

Trust must be earned continuously

Identity models built around static roles, persistent privileges, and occasional reviews are not sufficient to secure AI agents. For enterprises, the priority is clear:

  • Continuously understand identity risk across human, non-human, and AI identities.
  • Reduce standing privilege through access decisions informed by real-time context.
  • Establish trusted identities for AI agents and grant only the access they need, when they need it.

As AI agents become more capable and more deeply embedded in business operations, trust can no longer be treated as a one-time decision. It must be continuously established, evaluated, and enforced.

 

Protection from intelligent threats

Artificial intelligence is becoming part of everyday business operations. From automating routine tasks and analysing large volumes of information to supporting customer service, software development, and decision-making, AI presents significant opportunities for businesses of almost every size. The same technologies are also available to threat actors.

Cybercriminals can use AI to improve phishing campaigns, research potential targets, create convincing fraudulent communications, and impersonate real people. Attacks that once required considerable time, technical ability, or knowledge of an organisation can increasingly be created and adapted at speed.

This does not mean that AI has completely reinvented cybercrime. Phishing, social engineering, impersonation, and fraud have existed for decades. What AI changes is the potential speed, scale, and sophistication of these attacks.

For organisations, the challenge is therefore not simply deploying better cybersecurity technology. It is ensuring that employees can recognise increasingly convincing threats, that appropriate controls exist to verify important decisions, and that AI itself is used responsibly throughout the business.

How AI is being weaponised in the workplace

One of the clearest examples of AI-enabled cybercrime is phishing. In the past, many phishing emails contained obvious warning signs: poor spelling, unusual grammar, generic greetings, or language that did not quite fit the organisation supposedly sending the message. Generative AI can remove many of these clues.

A threat actor can combine publicly available information from company websites, professional networking platforms, social media, and other sources with generative AI to create highly personalised messages. An employee might receive an email referring to a genuine colleague, supplier, project, or event, written in professional language and apparently providing a perfectly reasonable explanation for the request being made.

The objective, however, remains familiar: persuade somebody to click a malicious link, disclose credentials, open an attachment, transfer money, or provide confidential information. AI can also make impersonation significantly more convincing.

Voice cloning can reproduce a person’s voice from relatively small samples of recorded speech, while deepfakes can create increasingly realistic video and imagery. These capabilities make business email compromise and executive impersonation more convincing.

Imagine receiving an urgent telephone call apparently from your Finance Director asking you to process an unusual payment. What happens when the caller not only knows the names of the people involved, but also sounds like the Finance Director? The same principle can extend to video calls, recorded messages, and apparently legitimate supporting documents. This creates an important new security reality: recognising someone’s voice, face, or writing style can no longer be treated as sufficient proof of identity.

AI can also accelerate reconnaissance. Attackers can use automated tools to collect and analyse large amounts of publicly available information, helping them understand organisational structures, identify senior employees, discover suppliers, and develop convincing scenarios for social engineering attacks. The underlying attacks may not be new, but AI can make them more believable, more personalised, and easier to conduct at scale.

Cybersecurity habits for an AI-enabled world

As attacks become more convincing, employees need to look beyond whether something appears legitimate. A useful principle is to verify the request, not just the person making it.

If somebody receives an unexpected request to change supplier bank details, for example, recognising the sender’s email address should not be enough. The change should be verified using an established process and, where appropriate, through a separate trusted communication channel.

The same principle applies to requests for passwords, sensitive information, confidential documents, or unusual financial transactions. Employees should also be alert to urgency. Social engineering frequently attempts to prevent people from thinking carefully: this payment must be made immediately; the Managing Director is waiting; the customer needs this now. AI has not changed that psychological technique. It has simply made the message delivering it more convincing.

Strong technical controls remain essential. Multi-Factor Authentication (MFA) can make stolen credentials harder to exploit. Email filtering, endpoint protection, access controls, and monitoring can help prevent, detect, and respond to malicious activity, but technical controls cannot anticipate every situation.

Employees should therefore know how to check suspicious links, handle unexpected attachments and QR codes, verify unusual requests, and report concerns quickly.

They should also understand that AI itself introduces information-security risks. Copying confidential business information, personal data, intellectual property, or customer information into an unapproved public AI service could result in that information being processed or stored outside the organisation’s control.

Cybersecurity awareness must therefore increasingly include not only defending against malicious uses of AI, but also using legitimate AI securely.

Building responsible AI into the organisation

Responsible AI use requires more than simply making AI tools available to employees. Businesses need to establish clear boundaries around how AI can be used, what information can be provided to AI systems and when human oversight is required.

An AI acceptable-use policy is an important starting point. Employees should understand which AI tools are approved, which uses are prohibited or restricted, and what types of information must never be entered into unapproved systems. These requirements need to be communicated clearly and reinforced through practical guidance. However, responsible AI extends beyond information security.

Businesses should consider data protection and privacy, intellectual property, confidentiality, accuracy, transparency, and potential bias. AI-generated information may appear authoritative while being inaccurate or incomplete. Where outputs could influence significant business decisions, they should be checked against reliable sources rather than simply accepted at face value.

Accountability also matters. If an AI system contributes to a decision, organisations need to establish who is responsible for that decision. Employees should know when they can rely on AI to assist their work and when outputs require additional validation, approval, or specialist review.

Third-party risk also needs consideration. An organisation may have robust internal controls but still introduce risk through suppliers whose products or services incorporate AI. Understanding where AI is being used within the supply chain, what information it processes and how that information is protected is increasingly important.

Established management-system approaches can help organisations manage these risks systematically. ISO/IEC 27001 provides a framework for managing information security risks, while ISO/IEC 42001 applies management-system principles specifically to the governance and responsible use of artificial intelligence.

The objective should not be to prevent innovation. It should be to create an environment where people can use AI productively while understanding the boundaries within which that use is safe, ethical, and accountable.

Turning employees into the first line of defence

Cybersecurity discussions have often described employees as the organisation’s “weakest link”. While there is some truth in that description, I increasingly question whether it is a helpful way to think about cybersecurity. People certainly make mistakes, and attackers deliberately exploit human behaviour. But employees are also uniquely capable of recognising when something does not make sense.

Technology may determine that an email originated from a legitimate account. An employee may recognise that the request contained within it is completely out of character. A video call may appear entirely legitimate. An employee may nevertheless question why a senior executive is suddenly asking them to bypass the organisation’s normal payment process.

Rather than treating employees as a cybersecurity weakness that needs to be controlled, organisations should recognise and develop their potential as an effective first line of defence. This requires more than annual awareness training.

Training needs to reflect the situations employees may now encounter. A finance team could be presented with a simulated voice call requesting an urgent transfer. HR employees could encounter an AI-generated recruitment scam. Procurement teams could practise responding to fraudulent supplier bank-detail changes. Employees using generative AI could be challenged to identify what information should and should not be provided to an external AI service.

These scenarios teach people how to make security decisions rather than simply asking them to remember a list of warning signs. Just as importantly, organisations need to create a culture in which employees are encouraged and supported in challenging unusual requests.

In an earlier article, “Deviation from the norm”, I discussed the importance of questioning requests that depart from established practices, particularly where urgency or apparently reasonable explanations are used to encourage compliance. AI has not changed that principle. It has made recognising those deviations, and having the confidence to challenge them, increasingly important.

An employee who delays a significant payment for five minutes while independently confirming the instruction is not being obstructive. They are applying an important security control. This becomes particularly important when requests appear to come from senior management. Attackers frequently exploit authority because employees may be reluctant to question somebody more senior.

Employees need to know that established security processes still apply when the request appears to come from the Chief Executive. Reporting processes also need to be simple and supportive. Employees who think they may have clicked a malicious link or disclosed information should feel able to report the incident immediately. The faster the security team knows about a potential compromise, the faster it can respond. Blaming people for genuine mistakes can have precisely the opposite effect, encouraging employees to remain silent when rapid reporting matters most.

The goal should be a workforce that is cautious without becoming fearful: people who understand the threats, know what controls are available, and feel confident enough to stop and verify something when it does not seem right.

From trust to verification

Artificial intelligence is changing not only how organisations operate, but also how criminals attempt to exploit them. The answer is not to become afraid of AI or to assume that every email, telephone call, or video meeting is fraudulent. Nor can organisations rely solely on cybersecurity technology to solve the problem. Businesses need layers of defence.

Technical security controls remain fundamental. Clear governance and responsible AI policies establish the boundaries within which technology can be used. Verification processes make impersonation and social engineering harder to exploit. Well-trained, confident employees provide something technology alone cannot always deliver: context and judgement.

In an AI-enabled workplace, seeing is no longer necessarily believing. Neither is hearing. Verification has therefore become increasingly important. Organisations need to move from trusting communications because they appear genuine towards confirming important actions through established processes.

Most importantly, businesses should recognise that their employees can be one of their strongest cybersecurity controls. Given the right training, processes, and organisational culture, they can provide an effective first line of defence against threats that are becoming increasingly difficult to recognise.

Audit evidence in the age of AI

For many years, auditors have relied on documents, records, signatures, financial transactions, and other forms of evidence with a reasonable expectation that the material presented to them is authentic. This does not mean auditors have simply accepted everything at face value. Professional scepticism has always required auditors to question evidence, consider its source and reliability, identify inconsistencies, and seek corroboration where necessary. Fraud, document manipulation, and falsification are certainly not new risks.

In my own experience of auditing, confidence in evidence has never come simply from what a document says or how convincing it appears. Understanding where evidence originated, how it was produced, and how it relates to other available evidence has always formed part of assessing whether it can reasonably be relied upon.

What is changing is the ease with which convincing evidence can be created. Advances in artificial intelligence mean that documents, correspondence, images, recordings, approvals, and entire datasets can now be generated artificially. The results can look authentic, internally consistent, and entirely plausible.

For auditors and audit teams, this introduces a fundamental question. It is no longer sufficient to consider what the evidence tells us. Auditors must also ask what gives them confidence that the evidence is authentic.

The changing nature of evidence

Traditional audit techniques have developed around assessing the sufficiency, appropriateness, and reliability of evidence.

An auditor might examine a document, identify its source, compare it with other records, confirm an approval, or trace a transaction through an organisation’s systems. Where evidence originates from an independent or authoritative source, greater confidence may generally be placed in its reliability.

Artificial intelligence does not make these principles obsolete. It does, however, change the environment in which they are applied.

Creating convincing false evidence once required time, specialist knowledge, and, in many cases, considerable effort. Generative AI has substantially reduced those barriers.

I recall one audit where I was provided with a spreadsheet containing Active Directory user account details. The information appeared unusually clean and tidy, so I asked to review the accounts directly within Active Directory. The two did not match. I subsequently discovered that the spreadsheet had been cleaned up as part of a review of leaver accounts, but the corresponding changes had not yet been made in Active Directory.

There was nothing inherently wrong with the spreadsheet, but it demonstrated an important distinction between information presented for review and the underlying source system. The reliability of evidence can depend as much upon its provenance as its appearance, a distinction that becomes increasingly important when AI can be used to create or transform information so convincingly.

One thing my auditing experience has taught me is that the absence of evidence does not necessarily mean that something has not happened. I have encountered situations where people genuinely believed that the evidence I was requesting did not exist, only for it to emerge later from another system, department, or individual. Equally, information provided in good faith may turn out to be inaccurate or incomplete. Neither situation necessarily involves deliberate deception, but both demonstrate why auditors need to look beyond the information initially presented to them.

A fabricated approval can resemble a genuine executive communication. A false document can adopt an organisation’s terminology, formatting, and writing style. Synthetic data can populate thousands of spreadsheet rows with realistic-looking transactions, values, dates, and descriptions.

Perhaps more importantly, AI could be used to generate not just an individual piece of false evidence, but an apparently coherent collection of supporting evidence.

A fabricated transaction could be accompanied by a purchase order, invoice, approval email, spreadsheet entry, meeting record, and supporting correspondence. Each item might appear to corroborate the others.

The traditional principle of corroboration therefore remains important, but auditors may increasingly need to consider whether apparently independent pieces of evidence genuinely originate from independent sources.

Professional scepticism in an AI world

Professional scepticism has always been central to effective auditing. Auditors are expected to maintain a questioning mind and critically assess the evidence presented to them. In an environment where convincing synthetic content can be created quickly, that mindset becomes even more important.

Professional scepticism may increasingly need to incorporate a form of digital scepticism. Looking back at my own audits, I recognise that I have already applied elements of this approach, although I would not necessarily have described it as digital scepticism at the time. Questioning where information came from, comparing it with source systems, and investigating something that did not appear quite right were simply part of the audit process.

This does not mean assuming that every document has been generated by AI or treating every dataset as potentially fraudulent. Such an approach would make effective auditing almost impossible. Instead, auditors may need to broaden the questions they ask about evidence, for example:

  • Where did the evidence originate?
  • Who, or what, created it?
  • Can its origin be independently established?
  • Has it been altered since it was created?
  • Does the evidence exist within the organisation’s source systems, or has it been produced specifically for the audit?
  • Can the underlying transaction or activity be verified through another source?

These questions move the auditor’s attention beyond the appearance of evidence and towards its provenance. In other words, the audit trail increasingly needs to demonstrate not only what happened, but also why the auditor can trust the evidence showing that it happened.

Can we identify synthetic evidence?

There may be indicators that warrant further investigation. These could include unusual statistical patterns, repeated structures, unexpected similarities between records, inconsistent metadata, improbable distributions, unexplained changes in writing style, or discrepancies between documents and the systems from which they supposedly originated.

Large datasets may present their own warning signs. Artificially generated data could contain patterns that appear random to a human reviewer but become apparent when the complete population is analysed. Values might be distributed too evenly, relationships between fields might be unusually consistent, or supposedly independent transactions might exhibit unexpected similarities.

However, auditors should be cautious about assuming that AI-generated material will always contain identifiable characteristics. Generative technologies will continue to improve, and methods designed to detect synthetic content will need to evolve alongside them. An apparent indicator of AI generation may also have an entirely legitimate explanation.

The objective should therefore not simply be to determine whether a document looks AI-generated. A stronger approach may be to focus on verification rather than detection. The more useful question is not whether an auditor can prove that something was generated by AI, but whether the evidence came from the source it claims to have come from.

Using technology to challenge technology

Artificial intelligence creates new risks for audit evidence, but technology can also form part of the response. Analytical tools allow auditors to examine far larger populations of information than would be practical through manual sampling alone.

Instead of reviewing a small number of individual transactions, auditors can increasingly analyse complete datasets, identify unusual relationships, compare information across systems, and highlight anomalies requiring further investigation. This could become particularly important when dealing with synthetic data.

Analysis might identify duplicate structures, improbable transaction sequences, unusual timing patterns, unexpected relationships between fields, or inconsistencies between operational systems and information supplied specifically for audit purposes. AI itself may also support this process by helping auditors identify patterns across large quantities of evidence that a human reviewer might reasonably overlook.

AI is contributing to the problem of evidence authenticity, but AI and advanced analytics may also become increasingly important tools for addressing that problem.

The challenge will be ensuring that auditors understand both the capabilities and the limitations of those tools. An auditor should not simply replace unquestioning trust in a document with unquestioning trust in an AI system analysing that document. Professional judgement remains essential.

Provenance may become as important as content

Historically, much of an auditor’s attention has understandably focused on the content of evidence, for example:

  • Does the invoice support the transaction?
  • Does the approval demonstrate appropriate authorisation?
  • Does the spreadsheet reconcile with the financial or operational system?
  • Does the record demonstrate that the required process was followed?

Increasingly, however, auditors may also need to understand the provenance of the evidence itself. A document retrieved directly from a controlled system may provide a different level of assurance from an identical document supplied as an email attachment.

A transaction independently extracted from the source system may provide greater confidence than a spreadsheet prepared specifically for an audit.

Evidence obtained directly by the auditor may carry different risks from evidence selected, compiled, or transformed before being presented by the auditee. None of these distinctions are entirely new to auditing, but AI changes their significance.

When convincing evidence can be generated cheaply, quickly, and at scale, understanding the route by which evidence reached the auditor becomes increasingly important.

The audit trail of the future

Traditionally, an audit trail allows an auditor to follow an activity or transaction through records and systems. In an AI-enabled environment, the audit trail may increasingly need to provide confidence in the origin and integrity of those records as well.

That could mean placing greater emphasis on source systems, access controls, document histories, system logs, digital signatures, timestamps, independently obtained evidence, and other mechanisms capable of establishing provenance. It could also mean reconsidering how evidence is requested.

Where possible, auditors may prefer to obtain information directly from authoritative systems rather than receiving documents that have been extracted, reformatted, or compiled specifically for the audit. This does not mean that every audit needs to become a digital forensic investigation. The response should remain proportionate to the risk.

However, where evidence is particularly significant to an audit conclusion, or where the circumstances create a heightened risk of manipulation, establishing authenticity may need to become an explicit part of the auditor’s evidence assessment.

A new dimension of professional scepticism

Artificial intelligence does not invalidate the fundamental principles of auditing. If anything, it reinforces them. Professional scepticism, corroboration, independence of evidence, reliability of sources, and professional judgement remain as important as ever. What is changing is the environment in which those principles must operate.

Auditors have always needed to consider whether evidence is reliable. The difference is that producing convincing false evidence is becoming easier, faster, and more scalable.

The challenge for the next generation of auditors may therefore not be learning how to recognise AI-generated evidence, but learning how to conduct assurance in an environment where authenticity can no longer be assumed from appearance alone.

When seeing is no longer believing, the strength of an audit conclusion will increasingly depend not simply on what the evidence says, but on how confidently the auditor can establish where that evidence came from, how it was created, and whether it can be trusted.