Security teams should treat phishing and vishing as identity and behavior problems, not just email problems. Reduce risk by correlating user behavior, access context, and current threat intelligence, then deliver targeted simulations and micro-training to high-risk groups. The goal is to intervene before a request is acted on, especially when urgency, authority, or voice impersonation is used to pressure employees.
Why This Matters for Security Teams
AI-generated email, chat, and voice content makes classic social engineering harder to spot because the attack no longer depends on poor grammar or obvious spoofing. The real risk is that AI can make a fraudulent request sound familiar, urgent, and context-aware enough to bypass human hesitation. Security teams should frame this as a control problem across identity, communications, and response workflows, not just an awareness problem. Guidance in the NIST Cybersecurity Framework 2.0 supports this broader view by linking awareness, detection, and response into a single operating model.
The practical failure is often not that employees cannot recognise a fake message, but that the request arrives through a legitimate channel with believable timing and enough detail to trigger action. Voice cloning adds a second layer of pressure because it can impersonate executives, help desks, or suppliers during a live conversation. That changes the defensive goal from spotting syntactic clues to validating intent, identity, and change in context before payment, credential reset, or data release occurs. In practice, many security teams encounter this only after a high-trust request has already been approved and the recovery work has begun, rather than through intentional prevention.
How It Works in Practice
Effective reduction starts by treating phishing and vishing as a workflow that can be interrupted at multiple points. Email gateways, call-back verification, secure request channels, and access controls should all reinforce one another. For example, a finance approval request should be routed through an authenticated internal system, while a password reset or MFA change should require a separate verification path with no reliance on caller ID or voice recognition alone. The MITRE ATT&CK Enterprise Matrix is useful here because it maps common techniques such as initial access, credential theft, and impersonation into defensive priorities.
Teams usually get better results when they combine three layers:
- Behavioral detection that looks for unusual request timing, device use, geo-location, and transaction context.
- Targeted simulations for employees who can trigger real business impact, such as finance, HR, service desk, and executives.
- Escalation playbooks that require out-of-band validation for sensitive actions like payment changes, payroll updates, or identity recovery.
Security operations should also integrate current threat intelligence so that simulation themes and detection rules reflect how attackers are actually operating. Public reporting such as Anthropic — first AI-orchestrated cyber espionage campaign report shows that AI can be used to scale reconnaissance, message generation, and operational follow-through, which makes fast pattern updates valuable. The broader defensive lens also aligns with CISA cyber threat advisories, especially when organisations need to translate public alerts into internal controls and user warnings. These controls tend to break down when approval happens through informal channels, because business pressure overrides the verification step.
Common Variations and Edge Cases
Tighter verification often increases friction for legitimate work, so organisations have to balance fast operations against stronger challenge procedures. That tradeoff becomes sharper in executive support, customer service, and supplier management, where urgent requests are common and attackers try to exploit speed.
Best practice is evolving for voice cloning specifically. There is no universal standard for relying on speech patterns, liveness cues, or voice print comparison as a primary control, so current guidance suggests using voice only as one factor within a broader verification workflow. If a call claims to come from a senior leader, the safer pattern is to verify the request through a known secondary channel, not to continue the conversation until it sounds convincing. Teams should also assume that attackers will adapt to training content, which is why simulation themes need regular refresh and should reflect live adversary tactics from MITRE ATLAS adversarial AI threat matrix.
Identity teams should pay special attention to reset and recovery paths, because those are high-value targets when attackers use AI-generated persuasion. Where help desk staff can override controls, the phishing problem becomes an identity assurance problem as much as a security awareness problem. The most resilient programs design for exception handling, not just normal-case verification.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK and MITRE ATLAS address the attack and risk surface, while NIST CSF 2.0, NIST AI RMF and NIST SP 800-53 Rev 5 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.AT | Security awareness and training reduce success of AI-driven phishing and vishing. |
| MITRE ATT&CK | T1566 | Phishing is the core delivery technique behind many AI-generated social engineering attacks. |
| MITRE ATLAS | AML.T0014 | Adversarial AI techniques include prompt-driven manipulation and deceptive content generation. |
| NIST AI RMF | GOVERN | AI governance helps organisations manage risks from AI-generated social engineering content. |
| NIST SP 800-53 Rev 5 | AT-2 | Awareness training is required to reduce human susceptibility to impersonation attempts. |
Track AI-enabled deception techniques and update simulations, detections, and playbooks accordingly.
Related resources from NHI Mgmt Group
- How should security teams reduce risk from AI agents and developer tools that use secrets locally?
- How should security teams use sensitive data discovery to reduce AI risk?
- How should security teams use DSPM to reduce oversharing risk in AI-enabled environments?
- How can security teams reduce risk from fast, queued AI content production?
