Gives survey evidence that frontier cyber-capable models have not yet raised consumer cyber incident rates (about 45% in September vs 46% in June). This tempers claims about immediate real-world harm.
Key takeaways · AI-distilled
Epoch and Ipsos found no statistically detectable rise in US adults reporting cyber incidents between June and September 2026. This finding does not rule out smaller changes that the surveys lacked power to detect.
Epoch and Ipsos's June and September cyber-incident polls ask about the preceding 12 months, so their recall periods overlap. They mostly sample different people and measure population-level reports rather than individual changes.
The Epoch-Ipsos survey does not identify AI-caused cyber incidents or estimate a particular model release's effect. Stable self-reported incident rates therefore do not establish that stronger AI had no cyber impact.
