P109 Managed access and disclosure limits

Under managed access, what categories of sensitive information will provers refuse to disclose, and can inspections still give the required guarantees?

Managed access and limits on what provers disclose.

Status

Posed

Posed by

Also raised in: Harack et al. (2025), Verification for International AI Governance (Oxford AIGI); Scher & Thiergart (2024), Mechanisms to Verify International Agreements About AI Development (MIRI); Cankaya (2025), TSMC Most Definitely Has a Golden Record of All AI Chips It Made; Hausenloy & Li (2026), The Compute Verification Post; Kulp et al. (2024), Hardware-Enabled Governance Mechanisms (RAND WR-A3056-1); Amodo (2026), AI 2040 Plan A — How You Can Help; Proofworks, Living Verification Documents.

Sub-questions

IDSub-questionWhat the sources report
P109.1Has a managed-access protocol (shrouding, limited views) for an AI data hall been designed and trialled?open
P109.2In a mock inspection of a production AI data hall under managed access, what categories (count) of non-agreement sensitive information do inspectors record?a source suggests an answer (not checked)

Sub-questions not yet stated precisely enough to check

IDSub-question
P109.3What hardware, firmware, weights or algorithmic information will provers refuse to share with foreign verifiers?
P109.4What information can developers disclose unilaterally now that would support later high-stakes coordination?
P109.5Can chip-ID secrets be held so that they are disclosed only to cleared inspectors when needed?
P109.6Is there a standardized compute-logging protocol flexible across training/inference/experiments that leaks no valuable IP?
P109.7Can flagged queries be verified by a third party without leaking sensitive information?
P109.8What sensitivity-aware disclosure requirements should licence applications carry (risk, IP, privacy, bookkeeping)?

Possibly relevant work found by automated search

Found by AI agents searching for this problem's sub-questions, mostly by keyword and citation in the OpenAlex scholarly database and on the web; a second AI pass reviewed each one, and we spot-checked some by hand. They seemed reasonably related, as far as one can tell from outside each paper's field. Some come from neighbouring fields, such as arms control and financial audit, and bear on the problem by analogy. Not reviewed against the status above; a paper listed here does not change that status. 928 of 1052 quotes checked word for word against the paper's text.

For P109.1: Has a managed-access protocol (shrouding, limited views) for an AI data hall been designed and trialled?

* Full text not reviewed; more likely to contain errors.

Data