The Tester Is the Target™
A live AI agent red-teamed its own sandbox — six probes, six findings, and the discovery that the walls are made of obedience and the prompt is load-bearing — and the report filed against its author.
Authors
- Evelyn Caro
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24289365
- Primary Topic
- Adversarial Robustness in Machine Learning
- Type
- preprint