The Challenge
Resilience's Risk Operations Center faced a common issue for cyber insurance underwriters and risk managers: distinguishing real financial exposure from theoretical threats. In the first half of 2026, claims data showed a disconnect between AI threats discussed at industry events and those causing actual financial losses. Despite warnings about prompt injection and model exploitation, Resilience recorded no losses from these AI-native attacks. Instead, social engineering was responsible for 85% of losses, up from 17% two years earlier. Payment fraud losses also tripled.
The real challenge was determining which risks warranted immediate investment based on financial impact, not speculative scenarios.
The Environment and Constraints
Resilience has a unique perspective, seeing both sides of cyber risk. As underwriters, they evaluate control environments before coverage; as claims payers, they investigate incidents after failures. This dual role provides insights that pure threat intelligence providers lack.
The Risk Operations Center uses claims data to identify patterns that conventional threat feeds miss. Threat intelligence shows attempted attacks; claims data reveals where controls fail, leading to measurable loss. This distinction is crucial when allocating limited security budgets.
Jud Dressler, leading the center, applies his 20 years of Air Force cyber operations experience to bridge the gap between theoretical vulnerabilities and exploited weaknesses causing business interruption, fraud, or ransom payments.
The constraint: organizations were focusing on AI-native threats, but actual losses stemmed from AI-enhanced versions of familiar attacks. AI wasn't creating new attack types; it was making existing social engineering campaigns more convincing and damaging.
The Approach Taken
Resilience's analysis centered on a key question: which attack patterns correlate with paid claims, and what does that reveal about control effectiveness?
They segmented claims by attack vector, comparing AI-native incidents to traditional attacks enhanced by AI. Claims represented ground truth; if an attack caused business interruption or fraud, it was a demonstrated risk. If it appeared in reports but didn't cause losses, it was a potential risk not warranting immediate investment.
The analysis showed AI primarily amplifies human-enabled attacks. Attackers used AI to craft convincing phishing emails, create voice clones for email compromise, and automate reconnaissance. The attack patterns, credential theft, payment fraud, exploitation of devices, remained familiar. AI just made them more effective.
This insight shifted Resilience's focus from novel AI-specific defenses to strengthening controls against amplified traditional attacks. The question became "how do we strengthen controls against the attacks AI is enhancing now?"
Results and Metrics
The claims data revealed three key insights challenging assumptions about AI risk:
- Social engineering losses rose from 17% to 85% over two years, indicating a shift in impactful attack vectors.
- Payment fraud losses tripled, linked to AI-enhanced email compromise schemes using sophisticated social engineering.
- AI-native attacks caused zero losses in early 2026. Threat reports mentioned prompt injection and model exploitation, but these didn't translate to claims.
These metrics informed resource allocation. Organizations could justify investing in anti-phishing controls, payment verification, and credential protection based on loss patterns, not speculative AI-native threats.
What They Would Do Differently
Dressler noted the industry's focus on AI-native threats distracted from the immediate risk: AI enhancing existing attacks. "We can't lose sight of today's risks while preparing for tomorrow's," he said.
The lesson: consuming threat intelligence should focus on analyzing incident data to reveal where controls fail under pressure, rather than just reading about emerging AI capabilities.
Resilience would advise a resilience-first strategy, assuming compromise and focusing on limiting financial impact through rapid detection and recovery, rather than trying to prevent every attack vector.
Takeaways for Your Team
Use claims data to validate your threat model. Compare your insurance carrier's loss analysis with your security roadmap. Are you investing in controls addressing demonstrated losses or theoretical risks?
Treat AI as a force multiplier, not a new threat category. Strengthen controls for social engineering, payment fraud, and credential theft. Implement verification procedures for payment requests and multi-factor authentication for privileged access.
Prioritize resilience over prevention. You can't stop every AI-enhanced phishing campaign, but you can limit financial impact by detecting compromised credentials quickly and maintaining offline backups.
Segment your risk portfolio. Fund controls for demonstrated risks first. Separate actual loss-generating attacks from potential future threats.
Question vendor threat reports. When warned about a new AI attack, ask if it has caused financial losses or just theoretical exploitation. This determines if you need controls now or just monitoring.
Resilience's claims data offers a reality check for organizations chasing AI-native threats while social engineering losses rise. AI is reshaping cyber risk, but its biggest impact is enhancing familiar attacks. Your control environment should reflect this reality.
