Preprint, version 5.3

Capability-Mediated Perimeters for Secure AI Agent Tool Execution

Conditional Non-Escalation Invariants and Empirical Evaluation Against Indirect Prompt Injection

Author

Rudraneel Das

DOI

10.5281/zenodo.22501491

License

CC BY 4.0, open access