TL;DR: AI-powered attacks now account for 41% of attacks targeting organisations, and Fingerprint’s survey of more than 300 leaders in SaaS, fintech, payment platforms, and banking shows average annual losses of $414,000, with more than a third reporting losses above $1 million. Privacy-first tools and regulations are also making legitimate users harder to distinguish from fraudsters, so fraud teams need stronger device intelligence and faster risk-based decisioning.
At a glance
What this is: This report says AI-powered fraud is now a mainstream operating condition, with 41% of attacks targeting organisations using AI and measurable losses showing up across SaaS, fintech, payments and banking.
Why it matters: Fraud and identity teams need to understand that privacy controls and AI-enabled abuse are now interacting, which changes how they assess user trust, investigate risk and balance friction against compliance.
By the numbers:
- 41% of attacks targeting organisations are now AI-powered.
- The survey of more than 300 leaders spans B2B SaaS, fintech, payment platforms and banking.
Context
AI-powered fraud is now a governance and operations problem, not just a detection problem. When attackers can automate reconnaissance, impersonation and workflow abuse, traditional fraud controls that rely on static signals or manual review become too slow to keep pace. The result is higher loss exposure, heavier investigation load and more friction for legitimate users.
The identity angle is strongest where fraud prevention depends on deciding whether a browser, device or session is trustworthy. Privacy tools such as VPNs, browser protections and regulation-driven data minimisation make that harder, which means teams need to separate user privacy from adversary anonymity without turning every transaction into a manual case.
Key questions
Q: How should fraud teams handle AI-powered attacks that look legitimate at first glance?
A: Treat the first decision as a trust decision, not a post-login investigation. Use device, session and behavioural context to score risk before account creation or transaction approval, then reserve manual review for edge cases that automation cannot classify with enough confidence.
Q: Why do privacy-first browsers and VPNs make fraud detection harder?
A: They remove or blur identifiers that fraud teams used to tie activity to a known device or session. That forces teams to depend on less invasive but more durable context, such as device reputation, interaction patterns and continuity across sessions, rather than assuming one identifier is enough.
Q: What are the signs that AI-fueled fraud is starting to affect fraud operations?
A: Common warning signs include a noticeable rise in spam, scam attempts, and account takeover activity, especially when the volume increases soon after a new AI tool becomes widely available. Teams may also see more variation in suspicious messages, more convincing impersonation, and more user reports of fraudulent contact. Those patterns suggest the attack surface is expanding faster than controls.
Q: When should organisations prioritise device intelligence over more manual review?
A: When privacy controls, cookie loss or session churn make legacy identifiers unreliable, device intelligence should take priority because it preserves continuity without depending on invasive data collection. Manual review should remain the exception for ambiguous high-risk cases, not the default.
Technical breakdown
How AI changes fraud economics
AI lowers the cost of producing believable scams, synthetic interactions and adaptive evasions. That shifts the attacker model from one-off events to scaled, repetitive pressure against onboarding, login and payment flows. Fraud teams then face a higher volume of low-signal events that must be sorted quickly, while the defenders’ detection stack is forced to distinguish automation, human users and blended attacks. In practice, the control problem moves from spotting a single bad actor to maintaining reliable trust signals across many sessions and channels.
Practical implication: fraud teams need stronger behavioural and device-level signals at decision time, not just post-event review.
Why privacy controls complicate identity signals
Privacy-first browsers, VPNs and regulation-driven data limits can remove or mask the very signals fraud teams have historically used to detect abuse. That does not make privacy controls wrong; it means the detection model must be rebuilt around less invasive but still durable context. Device intelligence, session consistency and risk history become more important when cookies are cleared, IP addresses shift or user-level identifiers are unavailable. The key challenge is preserving enough signal to govern access and transactions without over-collecting personal data.
Practical implication: teams should reassess which signals are essential for fraud decisions and which are only legacy convenience.
Manual triage becomes the bottleneck
When AI-powered attacks bypass traditional filters, the backlog moves into human review. That creates a second-order control failure: analysts spend more time confirming suspicious activity, which reduces their capacity to investigate the next wave of alerts. The article’s operational message is that fraud prevention is now partly a queue-management problem. If every ambiguous event becomes a manual case, cost rises and user experience degrades even when the final fraud loss remains contained.
Practical implication: reduce manual triage by pushing better trust scoring and automated pre-login decisioning upstream.
Threat narrative
Attacker objective: The attacker aims to scale fraudulent activity while avoiding detection long enough to convert trust into money or account access.
- Entry begins with AI-generated or AI-assisted fraud attempts that mimic legitimate behaviour across onboarding, login or transaction flows.
- The attacker exploits weak trust signals and privacy-obscured telemetry to evade traditional fraud checks and trigger manual review.
- The impact is financial loss, higher operational workload and slower decisioning for legitimate users and investigators.
Breaches seen in the wild
- SalesBleed Salesforce Agentforce 2026: Three fixed Agentforce flaws let poisoned web leads make AI agents leak CRM data with zero clicks and send phishing under the agent's identity.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
AI-powered fraud is now an identity governance problem, not only a fraud-scoring problem. Once attack generation becomes cheaper and more adaptive, the real control question shifts to how trust is established before a user or session is accepted. That makes device context, behavioural consistency and policy thresholds part of identity governance, not just fraud operations. Practitioners should treat the trust decision as a governance control, not a purely analytical output.
Privacy-first tooling creates a verification trust gap. The article shows a real tension between protecting users’ privacy and preserving enough signal to tell legitimate users from fraudsters. That gap matters because the controls that historically separated humans from automated abuse are increasingly suppressed, normalised or unavailable. The practical conclusion is that teams need privacy-respecting verification methods that still produce reliable risk context.
Manual triage latency: when AI-driven fraud outpaces automated detection, the review queue becomes a security control in its own right. Longer queues raise cost, slow customer decisions and create a wider window for repeat abuse. Organisations should therefore measure fraud operations as a time-bound control surface, not just a case-management function.
Device intelligence becomes a governance primitive for fraud teams. Persistent device recognition helps restore continuity when user-level signals are weakened by browsers, VPNs or cookie loss. That does not replace identity proofing or compliance obligations, but it gives teams a less invasive way to anchor risk decisions across sessions. Practitioners should align fraud policy, privacy requirements and device-level context as a single control design.
From our research library:
- U.S. fraud losses are projected to reach $40 billion by 2027.
What this signals
AI-powered fraud forces fraud teams to redesign the trust boundary. The issue is no longer whether a user is real in a static sense, but whether the session carries enough continuity to justify access, payment or onboarding. That pushes more decisioning to the edge of the journey, where device context and behavioural consistency matter more than after-the-fact review.
Device-level context is becoming a practical compromise between privacy and security. Teams do not need to abandon privacy principles to improve fraud outcomes. They do need governance that defines which signals are essential, which are excessive and how to preserve decision quality when user-level identifiers disappear.
For practitioners
- Strengthen pre-login trust scoring Move risk assessment earlier in the journey so suspicious sessions are scored before authentication or account creation, not after the fact.
- Reduce reliance on brittle identifiers Review which signals fail when cookies are cleared, IPs shift or VPNs are used, then replace them with more durable device and session context.
- Measure manual triage as a control bottleneck Track queue depth, investigation time and false-positive volume together so rising workload is treated as a security degradation, not just an ops issue.
- Separate privacy compliance from fraud telemetry design Document which signals are necessary for fraud decisions and minimise everything else so compliance does not accidentally remove the context needed for trust decisions.
Key takeaways
- AI-powered fraud is increasing both the scale and the realism of attacks, which makes trust decisions harder to automate safely.
- Privacy-first tools can weaken the signal fraud teams rely on, so legitimate users and fraudsters can look more similar at the point of decision.
- The most useful control shift is earlier risk scoring with durable device and session context, which reduces manual triage pressure and improves response speed.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
NIST SP 800-63 and NIST CSF 2.0 set the technical controls, while GDPR defines the regulatory obligations.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST SP 800-63 | SP 800-63B — Authentication | The article centers on distinguishing legitimate users from fraudsters at authentication time. |
| Recommendation — Use SP 800-63B to align authentication strength with the trust signal you actually have. | ||
| GDPR | Art.5 — Principles relating to processing of personal data | Privacy-first telemetry and fraud detection must stay within data minimisation and purpose limits. |
| Recommendation — Apply Article 5 to minimise fraud telemetry to what is necessary for the decision. | ||
| NIST CSF 2.0 | PR.AA-05 — Access Permissions, Entitlements and Authorizations | Fraud decisioning depends on whether access is granted to a session or user with enough confidence. |
| Recommendation — Tighten PR.AA-05 decisioning so access is granted only when trust signals are sufficient. | ||
Key terms
- Device Intelligence: Device intelligence is the practice of interpreting signals from a device to assess whether a session or transaction is likely legitimate. It goes beyond fingerprinting by combining device context with behavioural, identity, and payment evidence to support a risk decision.
- Privacy-First Fraud Detection: Privacy-first fraud detection is an approach that preserves user privacy while still collecting enough context to make a reliable risk decision. It relies on minimised, durable signals and governance around what is necessary for detection rather than broad surveillance.
- Manual Triage: Manual triage is the human-led process of sorting, validating, and prioritising security alerts one by one. It remains necessary for complex cases, but heavy dependence on it does not scale well. In practice, manual triage becomes a bottleneck when alert volume, staffing limits, and response expectations outgrow analyst capacity.
- Trust signal: Any cue that makes a person or system seem legitimate, such as a familiar name, known channel, authority marker, or expected behaviour. Fraud targets these signals directly, so security programmes must distinguish between recognition and proof.
Deepen your knowledge
NHI Mgmt Group covers identity security, NHI governance, and agentic AI through independent research, practitioner guides, and the NHI Foundation Level course. Explore nhimg.org for resources that connect identity governance to the broader security disciplines your programme depends on.
Published by the NHIMG editorial team on August 15, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org