Original title: Credentials That Carry Their Evidence: Credential design, dynamic tiering, and approved use when AI agents can do the work
Kumar Srivastava · August 2026
Artificial intelligence tools can now complete online classes for students, leaving employers unsure if a certificate was truly earned. Trying to prove a real human sat at the keyboard fails because someone can easily sit beside an artificial intelligence tool, and identity does not prove learning. To fix this, the author designs a new credential system. Under this plan, a certificate carries real evidence of a person's thinking, recorded during the learning process so outside parties can examine it. The setup pairs every certificate with an inspectable receipt and dynamically sorts skills into three tiers: unaided, directed, or attested. These tiers can upgrade or lapse over time. It also explicitly permits and measures tool use rather than banning it, proving value through inspectable proof rather than simple institutional signatures.
You are applying for a job that requires a cooking certificate. You took an online course and received a certificate, but the employer worries you might have just watched someone else cook. The usual solution is to have a proctor watch you, but that feels invasive and still does not capture your decision-making. In this paper, the certificate itself would include a detailed log of your cooking process, from ingredient selection to plating, so the employer can inspect your reasoning instead of just trusting the school's signature.
When an AI agent can complete a course for you, your diploma loses meaning. Employers know it might not reflect your learning. This paper shows how to make credentials carry evidence of the reasoning you actually did, so they stay trustworthy even when AI tools are used. It proposes ways to certify skills as unaided, directed, or attested, allowing credentials to be upgraded or to lapse as evidence accumulates.
The paper outlines what the proposed credential infrastructure requires but does not detail how to build it. It acknowledges what it does not solve, such as cases where reasoning cannot be captured or institutions refuse to adopt the new format. The author also discloses a conflict of interest, which may influence the proposal.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Credentials That Carry Their Evidence: Credential design, dynamic tiering, and approved use when AI agents can do the work — Kumar Srivastava — EdArXiv — doi:10.35542/osf.io/63rea_v1
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.