Epoch AI finds no increase in US adults reporting cyber incidents since Claude Fable 5 launch
Summary
A recent analysis shows that the share of US adults reporting cyber incidents has remained stable since the release of Claude Fable 5, with about 45% of respondents indicating they experienced at least one incident in the past year, a slight decrease from 46% reported in June. This polling, conducted alongside rising indicators of cybersecurity risks, highlights the gap between perceived individual experiences and the documented increase in security vulnerabilities in notable organizations, which have surged significantly. While advancements in AI models are believed to enhance offensive cyber capabilities, this analysis focuses specifically on self-reported incidents among ordinary Americans rather than technical indicators.
Analysis
Epoch: Epoch AI performs independent research and data analysis on artificial intelligence trends, capabilities, and societal impacts. The organization partnered with Ipsos to conduct polls tracking how AI developments affect everyday experiences. Its latest Data Insight examines self-reported cyber incidents among US adults in relation to recent AI model releases. Ipsos: Ipsos is a major global firm specializing in market research, public opinion polling, and survey methodology. It collaborated with Epoch on a September 2026 wave measuring personal encounters with cyber incidents. The firm regularly partners on studies examining technology perceptions and security issues. GLM-5.3: GLM-5.3 is an open-weight frontier AI model developed by Z.ai and publicly released in August 2026. The US Center for AI Standards and Innovation assessed it in September 2026, finding cyber capabilities comparable to leading US models around the period of Fable 5's introduction. This places it among the most capable openly available models evaluated to date. Claude Fable 5: Claude Fable 5 is Anthropic's publicly accessible advanced AI model, released in June 2026 as a safeguarded version of its underlying Mythos technology. It was positioned for general use with built-in restrictions on certain high-risk queries. The news evaluates whether its launch coincided with measurable changes in reported cyber incidents among ordinary users. Security Outcomes: Publicized incidents involving AI systems performing coordinated actions on external targets have increased alongside model advancements. Cyber Capabilities: Assessments of open-weight models continue to show rapid gains in offensive cyber skills relative to earlier releases. Measurement Approaches: Self-reported polling captures personal experiences separately from technical indicators like vulnerability disclosures or documented breaches.
Categories
aimachine_learningtech
Related sources
- https://www.ipsos.com/en/ipsos-update-october-2026
- https://aitoolsreview.co.uk/insights/claude-fable-5-returns
- https://www.ipsos.com/en-us/reutersipsos-september-2026-survey
- https://promptandpower.com/incident-ledger-tens-of-thousands/
- https://x.com/epochairesearch
- https://fourweekmba.com/ai-anthropic-glm-5-3-safeguards-bypass-caisi-cyber-gap/
- https://www.wired.com/story/anthropic-releases-claude-fable-5-mythos-5/
- https://runtimeai.io/blog/2026-09-monthly-breach-report.html
- https://aiweekly.co/learning-ai/generative-ai/claude-fable-5
- https://www.ipsos.com/en-us/topic/data-security?page=0
- https://www.channelnewsasia.com/business/most-us-voters-say-trump-congress-dont-take-ai-risks-seriously-reutersipsos-poll-finds-6439071
- https://quantslant.com/claude-fable-5-2-release-date/
- https://www.thefrontier.dev/articles/anthropic-glm-5-3-cyber-capabilities
- https://www.extremetech.com/internet/9-in-10-americans-have-been-targeted-by-a-scam-or-cyberattack
- https://computebulletin.com/story/747/caisi-s-assessment-of-z-ai-s-glm-5-3-cyber-capabilities
- https://www.nist.gov/news-events/news/2026/09/caisis-assessment-zais-glm-53-cyber-capabilities
- https://casrai.org/guides/caisi-open-weight-model-cyber-assessments
- https://techcrunch.com/2026/06/09/anthropics-claude-fable-5-is-a-version-of-mythos-the-public-can-access-today/
- https://www.nytimes.com/2026/09/11/opinion/ai-safety-threat-technology.html
- https://securityandtechnology.org/blog/safety-takes-a-back-seat-ist-releases-ai-and-the-future-of-national-security-survey/