Comment on FR Doc # 2026-12205
Rohan SharmaSupportIndividual
Summary: Rohan Sharma, an AI policy expert, supports the proposed GSA framework for safeguarding data within LLMs but proposes specific technical modifications. He argues for clarifying rules on exceptional human access for troubleshooting, establishing a 72-hour window for emergency cybersecurity notifications, and anchoring AI bias liability to objective benchmarks like the NIST AI RMF.
Rohan Sharma — Member, U.S. Technical Advisory Group to ISO/IEC on Artificial Intelligence; Member, ACM Technology Policy Committee; Aspen Institute Civic AI Leader; Author, AI and the Boardroom (Springer Nature, 2024).
Docket GSA-GSAR-2026-0331 / Notice-MVAC-2026-01
This submission addresses Questions 2, 3, and 4 of the notice and raises three targeted technical issues with GSAR 552.239-7001 as proposed.
Issue 1 (Paragraph (e)(4)(ii)): The clause does not establish conditions for exceptional authorized human access during troubleshooting, incident response, or RAG debugging, creating an operational compliance gap for Service Providers. I propose redline language authorizing time-bound, role-based, audited access for these scenarios.
Issue 2 (Paragraph (i)(4)): The "as soon as practicable" standard for emergency change notification is ambiguous and may delay deployment of critical cybersecurity patches in Federal enclaves. I propose a concrete 72-hour post-deployment notification window.
Issue 3 (Paragraph (j)(3)(ii)): Financial liability for non-compliance with "Unbiased AI Principles" is tied to subjective standards (historical accuracy, ideological dogma) that are not objectively measurable across LLM parameters. I propose anchoring liability to mutually agreed benchmarks or NIST AI RMF 1.0 to ensure reproducibility and commercial viability.
Full technical analysis with proposed redline language for each paragraph is provided in the attached document.
Submitted by Rohan Sharma in an individual capacity. All views expressed are my own and do not represent any institution or organization.