AI Cuts Phishing Costs 95%, Breaking Human-Vigilance Security Controls
Source
Aviation solved the vigilance problem. AI just gave security a worse oneCSO Online / StrongestLayer
What happened
A CSO Online analysis by StrongestLayer published on September 24, 2026 argues that AI has created a two-sided attack on human-dependent security controls. On one side, AI eliminates the traditional markers that trained workers use to identify phishing, such as poor grammar, odd formatting, and generic salutations. On the other, it floods workers with higher communication volumes, deepening the cognitive load problem. The article cites research by Bruce Schneier and colleagues showing AI reduces phishing campaign costs by over 95%, meaning sophisticated, personalized attacks are now economically routine. Drawing on the aviation vigilance decrement and the 2002 Ueberlingen mid-air collision, the piece argues that any control requiring sustained human attention will eventually fail under load, particularly for high-consequence, low-frequency actions like wire transfers and credential resets. The authors conclude that procedural controls that fire on the act itself, independent of whether the triggering communication appears legitimate, must now serve as the primary control layer.
Why it matters
- ·Controls that rely on a worker recognizing a fraudulent communication before approving a wire transfer or credential reset are now structurally weaker. The 95% cost reduction in phishing campaigns means organizations can no longer treat high-quality social engineering as rare or expensive to produce.
- ·Dual-authorization and out-of-band callback controls for high-consequence actions must be re-examined as primary, not supplementary, controls. Compliance programs that classify these as compensating controls rather than primary ones may be misaligned with the actual threat environment.
- ·Organizations in financial services, legal, and healthcare face the sharpest exposure because their high-consequence authorization actions, such as funds transfers, privileged access grants, and data releases, are the targets most attractive to attackers operating at AI-enabled scale.
Governance controls affected
What to do now
- ☐Audit every high-consequence authorization workflow (wire transfers, credential resets, privileged access grants) to determine whether human approval relies on evaluating the legitimacy of the triggering communication, and redesign any that do.
- ☐Require out-of-band verification (a separate phone call to a known number, a physical token confirmation, or a pre-registered callback) for all wire transfers and privileged access changes above defined thresholds, regardless of how the request was received.
- ☐Update your human oversight classification rationale logs (HOC-021 equivalent) to document explicitly why each high-consequence workflow either does or does not depend on a worker's ability to detect fraudulent communications.
- ☐Brief security awareness training designers on the structural argument in the Schneier and Heiding research: training workers to spot phishing is now insufficient as a primary control for high-stakes transactions, and messaging should reflect this shift.
- ☐Review your AI incident response playbook to confirm it covers social-engineering-enabled fraud as a category distinct from technical AI system failures, with specific escalation and notification paths.
What to watch next
Regulators in financial services are beginning to treat social-engineering losses as evidence of control deficiency rather than unforeseeable crime, a framing that could affect supervisory findings and enforcement. As AI-enabled fraud scales, guidance from bodies such as the Financial Stability Board, whose Sound Practices for Responsible Adoption of Artificial Intelligence addresses AI risk in financial contexts, may expand to address the adequacy of human-authorization controls specifically. Compliance teams should also monitor whether frameworks like the NIST Artificial Intelligence Risk Management Framework Playbook update their human oversight guidance to reflect the structural argument that vigilance-dependent controls are insufficient for high-frequency, low-cost AI-enabled attacks.
Stay ahead of stories like this
Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.