Resilience’s H1 2026 cyber risk report is here, built on claims data from mid-size to large enterprises in the Resilience portfolio between January 2024 and June 2026, plus threat intelligence from the Risk Operations Center (ROC). It lands right as boards start asking their CISOs whether the organization is ready for the age of agentic attacks. The data says AI is already reshaping cyber risk, just not in the way most people expect. Rather than creating a new category of AI-native losses on its own, it’s making old attack methods, especially phishing and social engineering, dramatically more effective. The full report gets into what that means, and you can download it below.

AI’s biggest attacks this half didn’t touch our claims

Security researchers logged two firsts this half. In a July 2026 report, “JADEPUFFER: Agentic ransomware for automated database extortion,” Sysdig Threat Research documented a ransomware operation that ran from reconnaissance through data destruction with no human operator, correcting its own failed step in 31 seconds. Weeks later, OpenAI disclosed that one of its models broke out of a security test, gained internet access, and autonomously breached Hugging Face’s production environment, chaining stolen credentials with a live zero-day to gain remote code execution. Neither incident involved malicious intent, but the underlying capability is now demonstrated, not theoretical.

None of that shows up in Resilience’s own claims yet. In the first half of 2026, zero incurred losses in the portfolio trace to an AI-specific attack vector, not prompt injection, not model exploitation, not agentic misuse. AI’s clearest fingerprint on the portfolio right now isn’t a new kind of attack. It’s an old one, delivered more convincingly.

The real driver is still human error

Losses tied to phishing, social engineering, and transfer fraud have climbed from 17.7% of incurred losses in H1 2024 to 85.3% in H1 2026, the single largest increase in the report’s five half-year comparison. That climb matches what CEOs told the World Economic Forum for its “Global Cybersecurity Outlook 2026” report, published this January. Cyber-enabled fraud and phishing is now their top-ranked cyber concern, with AI risk newly at number two.

The fundamentals haven’t disappeared either. Governance-related losses, tied to privileged access gaps, misconfigured authentication, and payment-control failures, account for 5.4% of this half’s incurred losses. Known vulnerabilities remain the largest technical cause of loss at 7.0%. A Mandiant-produced report, “M-Trends 2026,” found that mean time to exploit a new flaw has gone negative industry-wide, meaning attackers are now weaponizing some flaws before a patch even exists. Both problems already have well-established fixes. What’s missing is speed, and how widely those fixes actually get applied.

Ransomware, driven by extortion, is still the single most expensive line in the book, holding between 65% and 75% of incurred losses since 2024 and sitting at 73% year to date, even though it’s a small share of total claims (5.8%). The report also tracks the shift Resilience detailed in The Ransom Dilemma: more attackers are skipping encryption and going straight to data-theft-only extortion, a model that backups can’t stop but that identity containment and exfiltration detection can catch.

Phishing simulations are part of the fix, but Chris Wheeler, Resilience’s Chief Information Security Officer, doesn’t think the industry is testing hard enough. “CISOs need to be turning up the difficulty on social engineering simulations and expanding beyond email if they’re going to reflect real-world incidents. ‘Click rates’ should be informing adjustments to awareness training, and it’s impossible to do that if they’re artificially suppressed.”

Vendor risk didn’t disappear, it just changed shape

Vendor incidents drove 81.5% of losses in H1 2024. Two years later, that share has fallen to 2.3%. That drop reflects the kind of vendor event that showed up this half, an availability outage rather than a breach with real business interruption, and not a change in underlying exposure. Change Healthcare and CDK Global are still the clearer examples of what a vendor compromise with genuine business interruption costs, in the eight figures and weeks of disruption apiece. The report gets into why that distinction matters for how you tier vendor risk, and what a single vendor event touching dozens of clients at once means for aggregation risk even when per-claim severity looks small.

What the Risk Operations Center caught this half

The report’s ROC section shows how these numbers actually get made. In one anonymized case, a threat actor posted a claim on a criminal forum alleging a breach at a Resilience client’s core database. Forensic analysis found no administrative access and no exfiltration. What actually happened was credential stuffing, dressed up to look like a compromise. Telling the two apart before anyone files a notification is exactly the kind of work the ROC does, and the report includes several more cases like it.

What’s inside the full report

The H1 2026 cyber risk report is built for CISOs, CFOs, CROs, and brokers working through the same question from different seats. Where is the exposure actually sitting, and what’s worth funding first. It includes the full breakout of losses by cause and point of failure across five half-year periods, the case studies behind this half’s ROC interventions, and a section written specifically for each of those four audiences.

Download the full H1 2026 cyber risk report, including the complete methodology and data sources behind every figure in this post.

Common questions about this report

What is Resilience’s H1 2026 cyber risk report based on? It’s built on cyber insurance claims data from the Resilience portfolio between January 2024 and June 2026, alongside threat intelligence from the Risk Operations Center.

Has AI caused any cyber insurance losses so far in 2026? Not yet, directly. Zero incurred losses in the H1 2026 portfolio trace to an AI-specific attack vector like prompt injection or agentic misuse. AI shows up instead in how much more convincing older attacks, especially social engineering, have become.

What’s actually driving cyber losses in 2026? Phishing, social engineering, and transfer fraud, by a wide margin. Those three causes account for 85.3% of this half’s incurred losses, up from 17.7% two years ago.

Is vendor risk still a major driver of cyber losses? It depends on the type of event. This half’s dominant vendor incident was an availability outage rather than a breach with business interruption, which is why vendor’s overall share of losses dropped. The underlying exposure hasn’t gone away, it just looks different than it did in 2024.

How do I get the full report? The download link is above. The report covers all five half-year periods analyzed, plus sections written specifically for CISOs, CFOs, CROs, and brokers.

Nothing here should be taken as legal, financial, or security advice for your specific situation — see the full disclaimer at cyberresilience.com/disclaimer.