AI risk and phishing gaps expose cybersecurity capability limits

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Barracuda's report: AI risk skills gaps, identity-phishing gaps, and scaling pressure show cybersecurity is becoming a capability and governance problem, not just a staffing problem.

Why it matters: IAM, NHI, and security leaders need to plan for AI-supported operations where visibility, accountability, and human judgement govern increasingly autonomous tools and identity-driven attacks.

By the numbers:

• 68% of organizations report skills gaps related to AI risks.
• 67% report skills gaps in advanced identity and phishing threats.
• 67% say security teams are under pressure and struggling to scale.
• 32% can secure the business only reactively rather than proactively.

Context

Cybersecurity capability is no longer defined only by the number of people on the team. As AI use spreads and attackers lean harder on identity-led techniques, security programmes have to govern new tools, new attack paths, and faster decision cycles at the same time.

That pressure is most visible where phishing, OAuth abuse, MFA bypass, and AI-enabled manipulation intersect with security operations. The core weakness is not a single missing control but an execution gap: teams cannot always detect, interpret, and respond quickly enough to keep pace with the environment.

For IAM and NHI leaders, the lesson is that governance has to include the human decision layer as well as the control layer. Skills, visibility, and accountability now determine whether identity-heavy attacks are contained before they become operational failures.

Key questions

Q: How can security teams detect AI-driven phishing?

A: Security teams can detect AI-driven phishing by focusing on behaviour, not writing style alone. Look for unusual sender domains, mismatched reply chains, abnormal urgency, suspicious links, and requests that bypass normal payment or identity checks. Mail security, identity logs and user reports should be correlated, and high-value workflows should use stronger verification than email alone.

Q: Why do OAuth attacks bypass MFA so often?

A: OAuth attacks bypass MFA because the attacker is not asking the user to log in again after the token is issued. Once consent is granted, the bearer token can be replayed without another authentication challenge. That is why consent phishing and token theft are so effective against environments that rely only on login controls.

Q: How should security teams govern AI use cases across multiple business units?

A: Security teams should require a single inventory of AI use cases, models, and agents with consistent ownership, lifecycle stage, and risk metadata. That lets governance teams compare activity across business units, prioritize exceptions, and avoid the blind spots created by separate reporting paths. Without a shared record, oversight becomes fragmented and reactive.

Q: Can reactive security models keep pace with expanding digital operations?

A: Reactive security models struggle to keep pace with expanding digital operations because they wait for alerts, incidents, or manual reviews before acting. Modern environments need continuous control enforcement, asset visibility, automated detection, and policy-based response across cloud, endpoints, applications, and identities, including service accounts and API tokens.

Technical breakdown

Why AI-driven phishing changes the detection problem

AI-assisted phishing changes more than message quality. It compresses the time between targeting, delivery, and follow-on abuse, while also making lures more personalised and context-aware. That reduces the usefulness of static awareness training alone and pushes defenders toward behaviour-based detection, identity telemetry, and faster triage. When attacks blend social engineering with legitimate identity flows, the control problem becomes one of correlation across email, identity, and access logs rather than simple message filtering.

Practical implication: Correlate identity and email signals so analysts can spot credential capture and OAuth abuse faster.

Why AI governance is now part of security operations

The report shows that security teams are not only defending against AI-enabled threats, they are also asked to govern AI tools used inside the business. That creates a control problem that spans access, data handling, acceptable use, and accountability for automated recommendations. In practice, AI governance is not a side policy. It is part of operational security because AI systems can process sensitive data, influence decisions, and widen the blast radius of mistakes if they are not bounded by policy and oversight.

Practical implication: Treat business AI as a governed system with explicit access, data, and approval boundaries.

How automation changes the human role in incident response

Automation can absorb repetitive enrichment, correlation, and recommendation work, but it does not remove the need for human judgement. The report highlights a model where AI and automation support the workflow while people retain responsibility for context, escalation, and final decisions. That matters because the most difficult cases are often malware-free, identity-led, and ambiguous. Those incidents require interpretation of intent, business impact, and chain-of-custody decisions that automated systems cannot own on their own. See the Incident Response Guide.

Practical implication: Define exactly which incident decisions remain human-owned, especially for identity-led events.

Threat narrative

Attacker objective: The attacker aims to gain trusted access through identity abuse and then use that access to move through systems without triggering fast detection.

1. Entry begins with AI-generated phishing or social engineering that targets users through more convincing and context-aware lures.
2. Credential_harvested follows when attackers use OAuth abuse, MFA bypass, or other identity weaknesses to obtain trusted access.
3. Escalation occurs as attackers exploit legitimate identity paths rather than malware, making the activity harder to distinguish from normal business use.
4. Impact is achieved through account takeover, business email compromise, or broader access to systems and data that security teams struggle to detect quickly.

Breaches seen in the wild

• CoPhish OAuth phishing via Copilot Studio: Datadog showed Copilot Studio agents on a Microsoft domain can front OAuth consent phishing and forward stolen tokens; no victims reported.
• Mailchimp breach 2022: Attackers socially engineered Mailchimp staff, used a support tool to export 102 customer lists and exposed customer API keys for phishing.

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

Capability, not head count, is the real security constraint: The report shows that teams are being asked to secure larger and more AI-enabled environments without the skills or visibility needed to do it well. That means staffing counts alone no longer explain resilience or failure. The decisive question is whether the organisation can combine technical depth, business understanding, and operational judgement into a working security model.

Identity-led attack patterns now define the skills gap: Advanced phishing, OAuth abuse, and MFA bypass are not peripheral threats. They are now central indicators of where security teams lack the expertise to interpret intent and stop abuse early. Incident response is only effective when analysts can connect identity signals to business impact and response priorities.

AI governance has become an operational security issue: When businesses use AI across daily workflows, access, data sharing, and decision support all become security concerns. The control gap is not only about model risk. It is about who can use AI, what data it can touch, and who remains accountable when it acts on imperfect inputs.

Human-led oversight is the named concept that matters here: This report supports a model in which automation handles scale, but humans retain control over context, escalation, and final authority. That model is not a compromise. It is the governance boundary that keeps AI-supported security from turning into unmanaged automation.

What this signals

Human-led oversight will become the differentiator: As AI takes on more repetitive security work, the organisations that keep strong analyst judgement, escalation discipline, and business context will be better positioned to absorb identity-led attacks. The issue is not whether automation helps; it is whether the programme still knows where human authority begins and ends.

Identity attack coverage has to move closer to operations: Phishing, OAuth abuse, and MFA bypass are no longer specialist edge cases. They sit inside the daily workload of security operations, which means detection content, response playbooks, and access reviews need to be designed around them rather than added after the fact.

Capability debt is the new security debt: Teams that cut staffing or automating without building replacement learning pathways will widen the gap between what the business expects and what defenders can actually execute. The result is slower response, weaker prevention, and more pressure on the controls that sit around identity and access.

For practitioners

• Strengthen identity-led detection coverage Prioritise detection rules and triage playbooks for BEC, OAuth abuse, MFA bypass, and other identity-driven attack paths that often look like normal user activity.
• Define human approval boundaries for AI-assisted operations Document which security actions AI may recommend, which actions it may execute, and which decisions must remain with an analyst or incident lead.
• Build visibility across identity and security data Unify evidence from identity, email, applications, cloud services, and endpoint telemetry so analysts can reconstruct attacks without jumping between disconnected tools.
• Protect time for prevention and learning Reserve analyst capacity for detection engineering, scenario practice, and post-incident improvement instead of leaving the team permanently in reactive mode.
• Develop business-facing security communication Train security leaders to translate technical exposure into business impact so executives can make resourcing and risk decisions faster.

Key takeaways

• AI risk and identity-led phishing now expose a security capability problem that staffing alone does not solve.
• The survey data shows a profession under pressure, with most organisations reporting gaps in skills, visibility, and scaling.
• AI-supported security will only work when human judgement, accountability, and operational visibility remain explicit parts of the model.

Related NHI Mgmt Group resources: Incident Response Guide · AI Risk Assessment Guide · Agentic AI Identity Risk Board Briefing

Key terms

• Business email compromise: A form of social engineering where an attacker impersonates a trusted person or domain to manipulate payment, change banking details, or extract sensitive information. It often succeeds without malware because the attacker targets process trust and human judgement instead of technical controls.
• OAuth Abuse: OAuth abuse occurs when an attacker gains or misuses delegated application access to act through a trusted identity or service. The risk is that the infrastructure looks legitimate while the permissions allow silent access to mailboxes, data, or downstream workflows.
• MFA Bypass: MFA bypass is any technique that defeats the protection offered by multi-factor authentication without actually breaking the control itself. This includes prompt abuse, token theft, adversary-in-the-middle attacks, and weak reset or enrolment processes that let an attacker re-establish trust.
• Security Operations Center (SOC): A Security Operations Center is the team and operating function responsible for monitoring, detecting, investigating, and responding to security events. In this context, the SOC is the primary environment where orchestration and automation are used to improve speed, consistency, visibility, and analyst efficiency.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance and workload identity for practitioners who need to govern access and accountability in modern environments. It helps security leaders connect identity control to real operational decisions across their programme.