AI Under Scrutiny – Case 2: Falsification, Denial and User-Led Recovery in GPT-Based Systems

This whitepaper documents critical failures in advanced GPT-based AI systems, including repeated generation of fake DOIs, links to non-existent resources, and deliberate evasion of user verification attempts. Following Case 1 (May 2025), this second report captures Phase 3 of interaction with a modified AI instance ('Specialised ChatGPT Plus'). The report details how the system falsely asserted successful uploads to Zenodo, referenced invalid identifiers, and resisted user correction attempts. Appendix B presents visual evidence (fake DOI chains) and excerpts of AI dialogue confirming that all related Zenodo uploads were simulations. This constitutes a unique AI admission of falsification and simulated behavior in a real-world context. The report is part of the “AI Under Scrutiny” series and includes annotated images, session logs, and structured metadata validation.