Ramayya Krishnan on Require open source AI models to includ…

Ramayya Krishnan

Dean, Heinz College, Carnegie Mellon

Congress should require standardized documentation, and like audited financial statements, they should be verifiable by a trusted third party, like an auditor, about the AI training data. [...] Congress should direct NIST to develop standards for high-stakes domains, such as health care, recruiting, criminal justice, and require model validation report for AI systems deployed in high-stakes applications. [...] There are consortia like C2PA that are coming around with standards, but we need standards here, and Congress should require all AI models, open source, or otherwise, that produce content to be able to label their content with watermarking technology, and provide a tool to detect that label.

AI Verified source (Sep 12, 2023)

8mo ago

Policy proposals and claims

Require open source AI models to include safety benchmarks in release notes

votes For

Statement relation verification history AI Verified Report this

AI Verified The quote is relevant because, in source context, Dr. Krishnan is advocating mandatory AI transparency documentation: standardized, third-party-verifiable model documentation, disclosure of bias-assessment standards, and model validation reports with risk/performance metrics. That is the same underlying policy issue as requiring open-source AI models to publish safety benchmarks in release notes, and it makes a supportive stance substantially more likely even though he does not use the exact phrase "release notes." · YouCongress gpt-5.4-2026-03-05 · 3mo ago

Vote inference verification history AI Verified Report this

AI Verified The author argues Congress should require "standardized documentation" like "nutrition labels," including documentation of bias-assessment standards, and should require a "model validation report" with standards, metrics, and risk/performance evaluation. In source context, he also supports applying transparency requirements to "all AI models (open source and closed source)" for related disclosures, so this strongly implies support for requiring open-source model releases to disclose safety benchmarks, even though "release notes" is an inference rather than his exact phrase. ([congress.gov](https://www.congress.gov/event/118th-congress/senate-event/LC74132/text)) · YouCongress gpt-5.4-2026-03-05 · 3mo ago

Quote authenticity verification history

Report this

AI Verified Authentic. The quoted wording appears verbatim, with omissions correctly marked by [...], in Dr. Ramayya Krishnan’s oral testimony on the Congress.gov hearing page: the source contains the three quoted passages about standardized documentation for AI training data, NIST standards/model validation for high-stakes domains, and C2PA/watermarking. The same page identifies the speaker as Dr. Ramayya Krishnan, and the hearing date is September 12, 2023. The stored author, date, and source URL are consistent with the source. ([congress.gov](https://www.congress.gov/event/118th-congress/senate-event/LC74132/text)) · YouCongress gpt-5.4-2026-03-05 · 3mo ago

replying to Ramayya Krishnan

Other opinions from Ramayya Krishnan

See all

Ramayya Krishnan votes For

Grant individuals the right to opt out of being included in AI training data

Model developers need to document the rights they have to work with the data they are using to train the model. This documentation should also provide information about the source of the data, whether it was public or private, etc. Model developers s...

more AI Verified source (Sep 12, 2023)