Answers are grounded in this paper and cite their sources.
Protect the agent while it runs. Prove how it behaves before it ships. Account for what each correct result actually cost. That is the whole body of work, in one sentence. This write-up presents Karmendra Pandey's connected body of work on trustworthy AI agents under three headings — Protect, Prove, and Account. Protect covers the Argus runtime security layer that governs an agent while it runs. Prove covers two benchmarks: Argus-Bench, which audits safety and performance, and PersistAIBench, which audits memory over time. Account covers cost and accountability tooling, including a cost method that prices a correct result instead of a token. The write-up also covers scanners and browser tools that extend the same checks to the tools an agent connects to and the prompts a person types, and a working demo, Argus Chat, that runs them together in a single session. It ends with a directory of links and DOIs for the related works.
No takes yet. Share an insight, caveat, or question.
Karmendra Pandey (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: