AI-enabled fraud is reshaping cybercrime losses and identity trust

At a glance

What this is: This analysis says AI is making proven cybercrime tactics more effective, with fraud, impersonation, and BEC driving most losses.

Why it matters: It matters because IAM, NHI, and human identity programmes all depend on trust signals that AI can now imitate, scale, and sustain across workflows.

By the numbers:

• The FBI’s 2025 IC3 report recorded $20.877B in cybercrime losses, a 26% YoY increase.
• AI cybercrime led to $893M in losses from 22,364 complaints.
• Business email compromise caused $3B in 2025 losses.
• Government impersonation caused $797M in losses.

👉 Read Abnormal AI's analysis of AI-driven cybercrime and impersonation risk

Context

AI-enabled fraud is a governance problem as much as a detection problem. When attackers can mimic executive tone, vendor context, and timing at scale, the organisation’s trust model becomes the target rather than the payload. For identity teams, that means the boundary between human identity assurance, workflow control, and messaging integrity is now part of the same control surface.

The article’s central claim is that AI has not replaced familiar fraud techniques. It has improved their execution quality, persistence, and reach, especially in business email compromise and impersonation. That shifts the question from whether a message looks suspicious to whether the interaction fits the normal identity and communication pattern of the business.

Key questions

Q: How should security teams respond when AI makes business email compromise harder to spot?

A: Teams should move beyond message inspection and verify the requester, the channel, and the business context before allowing action. AI makes tone and wording unreliable signals, so the control point becomes workflow validation, out-of-band confirmation, and monitoring for abnormal approval patterns across finance, executive, and supplier interactions.

Q: Why do AI-enabled impersonation attacks create a human identity governance problem?

A: Because the attack succeeds by manipulating trust decisions made by people inside legitimate processes. When AI can replicate executive tone, vendor context, and urgency, the issue is no longer only malicious content. It becomes a governance problem for approvals, accountability, and the identity signals that support those decisions.

Q: What breaks when fraud detection relies only on known-bad indicators?

A: Known-bad indicators miss attacks that are newly generated, context-specific, and conversational. AI removes many of the cues that older filters depend on, so organisations can no longer assume suspicious language or formatting will be present. Detection must shift toward behaviour, relationship patterns, and approval anomalies.

Q: How can organisations reduce BEC risk without slowing legitimate work?

A: Use risk-based verification for high-value transactions, privileged changes, and sensitive requests rather than applying the same friction everywhere. The goal is to make high-impact actions harder to fake while keeping routine work efficient. That means stronger validation where trust has financial consequence.

Technical breakdown

How AI improves business email compromise execution

Business email compromise succeeds when an attacker can sound like a legitimate actor and place the request inside a real workflow. AI raises that success rate by generating messages with the right tone, vocabulary, and context, then adapting follow-up messages to maintain pressure. The technical change is not only better text generation. It is interaction sequencing across email, voice, and chat channels. That makes the attack less dependent on obvious grammatical clues and more dependent on behavioural plausibility. In practice, the defender is no longer comparing a single message to a known bad pattern, but evaluating whether the request chain resembles an established identity and approval path.

Practical implication: build controls that validate the requester, the channel, and the approval path before financial or privileged action proceeds.

Why impersonation scales faster when AI can preserve context

Impersonation has always depended on context. AI now lets attackers preserve that context across many targets without losing consistency in role, tone, or operational detail. That matters because trust is often built on small cues such as naming a real project, referencing a real manager, or mirroring internal communication style. With AI, those cues can be generated repeatedly and adapted per target. The result is scalable social engineering that no longer relies on one highly crafted message. Instead, it creates a believable identity frame that can be reused across executives, vendors, and employees with limited attacker effort.

Practical implication: strengthen human verification steps for high-risk requests and treat identity context as a control, not just message content.

Why persistent fraud campaigns are harder to stop than one-off phishing

The report highlights a shift from isolated attempts to sustained campaigns. Persistent fraud works because each interaction reinforces credibility, letting the attacker refine timing and pressure after every reply. AI reduces the cost of maintaining that sequence, so the campaign can continue long enough to align with normal business rhythms such as payroll, invoicing, or urgent payment requests. That creates a structural gap for controls that only look for single-event anomalies. The problem is not merely malicious content. It is the sustained manipulation of expectation over time, which makes the attack blend into routine work.

Practical implication: monitor multi-step conversation patterns and stop relying on one-message detection for fraud containment.

Threat narrative

Attacker objective: The attacker wants to convert manufactured trust into money movement, credential disclosure, or downstream access that can be monetised.

1. Entry occurs through a convincing AI-generated impersonation message or voice call that matches an executive, vendor, or government contact.
2. Escalation happens when the attacker sustains the conversation, reinforces legitimacy, and steers the target into approving payment or sharing sensitive information.
3. Impact is financial loss, reputational damage, or further compromise after the victim acts on the fraudulent request.

Breaches seen in the wild

• Cisco DevHub NHI breach — IntelBroker exploited exposed Cisco credentials, API tokens and keys in DevHub.
• DeepSeek breach — DeepSeek breach exposed 1M+ log lines and sensitive secret keys.

Read our 52 NHI Breaches Analysis report for a comprehensive view of breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI-driven fraud is now an identity problem, not just a content problem. The report shows attackers winning by manipulating trust signals that sit around the identity layer, including tone, context, and timing. That means human identity assurance and workflow validation are now part of fraud defence, not separate disciplines. Security teams should treat behavioural trust as a governed control surface, not a soft signal.

Persistent impersonation is the named concept practitioners need to track. The article’s real warning is not single-message phishing, but repeated, context-aware engagement that survives across multiple interactions. Once fraud becomes a sustained conversation, legacy signature-based controls lose their edge because the attack is designed to look ordinary over time. Practitioners should recognise that the unit of attack is now the relationship, not the message.

Static detection assumptions fail when attackers can manufacture believable normality at machine speed. The organisation assumes suspicious content will stand out, but AI can now produce requests that fit internal language and business rhythm. That breaks the old premise that message review alone can reliably separate legitimate from fraudulent activity. The implication is that identity, behaviour, and approval context must be analysed together.

Cyber-enabled fraud exposes a governance gap across human and machine-mediated trust. The article shows that the same attack family can target executives, finance teams, vendors, and support workflows with minimal variation. That creates a cross-domain problem for IAM, PAM, and awareness programmes because the attacker is exploiting human decisions through digitally mediated identity events. Practitioners need to align trust verification across all high-risk approval paths.

AI fraud pressure will keep shifting security budgets toward behavioural control planes. Once attackers can scale impersonation cheaply, the differentiator becomes how well an organisation understands normal relationships and approval patterns. That favours identity-linked telemetry, communication baselines, and transaction verification over content inspection alone. Teams that ignore this shift will keep funding the wrong layer of defence.

From our research:

• 43% of security professionals are concerned about AI systems learning and reproducing sensitive information patterns from codebases, according to The State of Secrets in AppSec.
• Organisations maintain an average of 6 distinct secrets manager instances, which fragments control and complicates consistent governance across environments.
• For the lifecycle angle, see Ultimate Guide to NHIs , Lifecycle Processes for Managing NHIs for how governance models change when identities and secrets span more than one control plane.

What this signals

Persistent impersonation: security teams should expect fraud to arrive as a sequence of believable interactions rather than a single malicious message. That changes the programme objective from spotting bad content to validating the legitimacy of a relationship over time, which is a stronger fit for identity-led detection and transaction assurance.

The practical signal for IAM and fraud teams is that approval-path integrity now matters as much as authentication strength. When attackers can fabricate executive context at scale, controls that only govern sign-in events will miss the moment where trust becomes money movement or data disclosure.

The risk posture should also be read alongside The State of Secrets in AppSec, where 43% of security professionals worry that AI systems may learn and reproduce sensitive patterns from codebases. That concern reinforces a broader pattern: AI turns internal context into attacker fuel if governance does not constrain what can be learned, reused, and acted upon.

For practitioners

• Tighten verification on high-risk approval paths Require out-of-band verification for payment, payroll, vendor bank detail, and privileged access changes, with a verified second channel before execution. Link the approval workflow to the named requester and the expected business context.
• Baseline communication patterns by role Model normal sender, recipient, timing, and escalation patterns for executives, finance teams, and suppliers so deviations can trigger review before action is taken.
• Rework fraud controls around conversation sequences Detect multi-step interaction patterns, not just suspicious messages, because AI-assisted fraud often starts with a plausible opening and escalates through follow-up pressure.
• Align identity teams with fraud operations Bring IAM, PAM, SOC, and finance controls into one response path for BEC and impersonation cases so identity signals can block transactions before money moves.

Key takeaways

• AI is amplifying fraud by making impersonation more believable, more persistent, and easier to scale across business workflows.
• The scale of the problem is already material, with the FBI tying $893 million in losses to AI-related complaints and $3 billion to BEC in 2025.
• Security teams should shift from content-only detection to identity, behaviour, and approval-path verification before high-risk actions execute.

Key terms

• Business Email Compromise: Business Email Compromise is a fraud pattern where attackers impersonate a trusted person or supplier to trick an organisation into moving money or sharing sensitive information. The attack succeeds by abusing business workflows and human trust, not by exploiting a software vulnerability.
• Impersonation attack: An impersonation attack is any attempt to present a false identity that the target is likely to trust. In modern environments this can span email, voice, chat, and document workflows, making identity context and approval discipline as important as technical email filtering.
• Behavioural baseline: A behavioural baseline is the expected pattern of communication, approval, or access activity for a person, role, or system. Security teams use it to detect deviations that suggest fraud, account takeover, or manipulated trust, especially when content-based detection is no longer reliable.
• Approval-path integrity: Approval-path integrity is the assurance that a request follows the right people, channels, and checks before a sensitive action is taken. It matters because AI-generated fraud often works by making an invalid request look normal inside an otherwise legitimate workflow.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity security are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are responsible for identity security strategy or NHI governance in your organisation, it is worth exploring.

This post draws on content published by Abnormal AI: FBI IC3 2025 cybercrime report analysis and AI-driven fraud findings. Read the original.