Phishing threat trends show AI-driven evasion is scaling fast

At a glance

What this is: Knowbe4’s latest phishing threat trends report says AI is helping attackers scale polymorphic phishing and evade Microsoft native security and secure email gateways more effectively.

Why it matters: This matters because phishing defence now has clear identity implications, with credential theft, ransomware delivery, and trust abuse outpacing email-only controls.

By the numbers:

• Cybercriminals have achieved a 47% increase in attacks evading Microsoft’s native security and secure email gateways.
• 92% of polymorphic attacks utilize AI to achieve unprecedented scale.

Context

Phishing remains effective because it targets human trust and identity controls rather than only technical perimeter defences. When attackers can adapt language, timing, and delivery at machine speed, message filtering becomes a partial control instead of a complete one. The identity implication is direct: every successful lure is an authentication and authorisation problem as much as an email problem.

The report frames a familiar pattern in a more aggressive form, with AI helping attackers revive old techniques rather than invent entirely new ones. That makes the governance challenge broader than spam reduction. Security teams now have to treat phishing as a cross-domain issue involving identity assurance, response speed, behavioural detection, and account hardening.

Key questions

Q: What breaks when phishing campaigns are generated and iterated by AI?

A: What breaks is the assumption that phishing is slow, manual, and easy to profile. AI lowers the cost of iteration, so attackers can test more messages, refine targeting faster, and scale successful lures quickly. Defenders lose time unless they can detect the downstream identity effects rapidly.

Q: Why do AI-assisted phishing and credential theft increase ransomware risk?

A: AI can improve targeting, language quality, and volume, which raises the success rate of social engineering and credential collection. The practical effect is more valid-account access at the front of the attack chain, so the defender has less time to detect abuse before the attacker reaches remote access, cloud console, or privilege escalation paths.

Q: What are the signs that email-only defences are no longer enough?

A: The clearest signs are rising message evasion, repeated user-reported phish that pass gateway controls, and suspicious account activity after delivery. If attackers can keep reaching inboxes with slightly different lures, the control set is too narrow and needs identity and behavioural layers.

Q: When should organisations prioritise phishing resistant access over more awareness training?

A: Organisations should prioritise phishing resistant access when the environment already has SSO, MFA, and frequent remote access, yet attacks still succeed. That pattern shows the weak point is authentication assurance, not user knowledge. Training remains useful, but it should support controls that reduce the chance of stolen credentials becoming usable access.

Technical breakdown

Polymorphic phishing and AI scale

Polymorphic phishing changes the content, structure, or delivery pattern of a lure so defenders cannot rely on static signatures or simple similarity checks. AI accelerates that variation by generating many plausible message versions, which raises the cost of manual review and makes conventional filtering less durable. The result is not just more phishing, but phishing that is harder to cluster, tune, and suppress at the gateway layer. Practical implication: move detection logic beyond static content matching and into behavioural, identity, and campaign-level analysis.

Practical implication: tune controls for pattern drift, not just known-bad indicators.

Why secure email gateways miss more attacks

Secure email gateways are built to inspect messages, links, attachments, and sender signals before delivery. That model struggles when attacks are distributed across channels, vary rapidly, or rely on trusted infrastructure and realistic conversational content. The article’s 47% evasion figure points to a control mismatch: defenders are trying to block adaptive content with controls that assume repeatable content. Practical implication: treat the email gateway as one layer in a broader detection chain, not the primary decision point.

Practical implication: add downstream telemetry and user-risk signals to reduce reliance on inbox inspection alone.

Ransomware delivery through phishing

Phishing remains a reliable entry path for ransomware because it can still deliver initial access, credential theft, or malicious execution before broader controls intervene. AI does not change the attack goal, but it can improve lure realism and increase the probability that a user or workflow will trust the payload. Once that initial trust boundary is crossed, the campaign can move from account compromise to encryption or extortion. Practical implication: identity protection and rapid containment matter before ransomware payload delivery becomes an operational event.

Practical implication: instrument identity, endpoint, and response controls around the first compromise, not only the final payload.

Threat narrative

Attacker objective: The attacker objective is to convert convincing phishing into reliable credential access or ransomware entry while staying ahead of inbox and gateway detection.

1. Entry occurs through AI-assisted phishing messages that are harder to distinguish from legitimate business communication.
2. Credential harvesting or malicious interaction follows when recipients trust the lure and expose account access or execute a payload.
3. Escalation then uses the compromised identity or foothold to support ransomware delivery, persistence, or broader campaign movement.
4. Impact is measured in account compromise, extortion, operational disruption, and reduced confidence in email-based trust signals.

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

AI-driven phishing is now an identity governance problem, not just an email-security problem. The report shows attackers using AI to scale lures, evade filtering, and keep phishing economically viable. That means the control boundary has shifted from message reputation to trust validation, account protection, and response speed. Practitioners should treat every successful phish as an identity event that can cascade into credential abuse.

Phishing resilience now depends on detection diversity. Static message rules, segmentation alone, and inbox controls cannot keep pace with polymorphic campaigns. Behavioural analytics, conditional access, MFA hardening, and user-reporting workflows all become more valuable when attackers continuously change the lure surface. The governance lesson is that a single choke point is no longer enough.

AI has lowered the cost of precision social engineering. That changes how security teams should think about awareness, simulations, and policy enforcement. The best programmes will measure how quickly they can identify, contain, and invalidate compromised sessions after a lure succeeds. In practice, that means shifting investment toward rapid identity containment and continuous monitoring.

Ransomware resurgence through phishing shows that old attack paths remain effective when attackers can automate variation. The named concept here is phishing adaptation debt, meaning the gap between a defender’s fixed controls and an attacker’s ability to mutate delivery fast enough to stay effective. When that gap widens, the organisation accumulates preventable exposure. Practitioners should close the debt at the control layer, not just in user training.

From our research library:

• The IBM/Ponemon 2025 Cost of a Data Breach Report found that phishing-initiated breaches cost an average of $4.8M each.

What this signals

Phishing adaptation debt: attackers are mutating lures faster than many organisations can update email controls, which means the real defence window has moved to identity enforcement and response. Security teams should measure whether they can invalidate a suspicious session before the attacker can reuse it.

The practical consequence for programmes is clear: inbox controls still matter, but they no longer define resilience on their own. Teams that align phishing response with conditional access, token revocation, and behavioural detection will contain more attacks before they become account compromise or ransomware staging.

For practitioners

• Harden identity controls for phished accounts Require phishing-resistant MFA for high-risk users and monitor for impossible travel, token abuse, and anomalous consent events after suspicious messages are reported.
• Add behavioural detections to inbox filtering Correlate sender reputation, message intent, and post-delivery user behaviour so campaign variation does not bypass static email rules.
• Shorten containment steps for credential theft Automate session revocation, password reset, and token invalidation for accounts exposed through phishing to limit lateral movement and ransomware staging.
• Test phishing resilience beyond awareness training Run simulations that include polymorphic lures, non-email delivery channels, and business-process impersonation so control gaps are visible before attackers find them.

Key takeaways

• AI is making phishing campaigns more adaptive, which weakens controls that depend on fixed patterns or static signatures.
• The article’s scale signals a control gap between message inspection and actual identity protection, especially when compromised sessions can be reused quickly.
• Teams should respond by hardening authentication, widening behavioural detection, and automating containment for credential theft events.

Key terms

• Polymorphic phishing: Phishing that changes content, structure, or delivery details so the same campaign appears different across messages. The goal is to defeat signature-based detection and increase the chance that at least one variant reaches a user and produces a credential, consent, or payload interaction.
• Phishing Detection Evasion: Phishing detection evasion is the set of methods attackers use to avoid being seen by traditional security controls. It includes non-email delivery channels, link camouflage, bot checks, obfuscation, anti-analysis, and MFA bypass. The practical challenge is not just blocking phishing, but detecting abuse before credentials or sessions are compromised.
• Identity Hardening: The practice of reducing identity-related risk by strengthening authentication, shrinking access scope, and improving lifecycle governance. It applies across human, service, and vendor identities when access is critical enough that weak authentication or standing privilege can materially raise exposure.
• Session revocation: The ability to invalidate active sessions so access ends immediately instead of waiting for tokens or browser state to expire. For identity governance, this is the control that determines whether authentication still matters after a compromise is detected.

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. It is suitable for practitioners who need to connect identity controls to broader security operations and risk governance.