- Verified, via search, the actual content of China's Interim Measures for Generative AI Services (2023, Art. 17 algorithm filing/security assessment), the 2022 Algorithm-Recommendation Provisions, PIPL Art. 38 cross-border data rules, and the 2023 Deep Synthesis content/labeling rules — not assumed from general knowledge.
- Verified the current (2026) diplomatic state: the May 2024 Geneva AI dialogue, the May 2026 Trump–Xi Beijing agreement to a further AI dialogue, China's 2026-08-31 stated preconditions for a September 2026 meeting, and China's own July 2026 launch of the rival World AI Cooperation Organization (WAICO), all from independent news coverage, not invented for this analysis.
- Verified that China's GM/T 0015 SM2 certificate standard is a real, published national cryptography standard that embeds identity attributes via custom OID extensions, and confirmed c=US's own demonstration certificate profile (
research/openssl with oid.txt) does the same thing structurally, under its own placeholder OID arc and ordinary Western PKI algorithms. - Wrote every factor description and every rubric level Jev was asked to score against.
Cross-border follow-up · regulation & shared control
Two narrower questions: is the accord compatible with Chinese AI law, and could c=US support shared control?
c=US's own demonstration certificate profile embeds custom operational attributes directly into an mTLS client certificate using a private-enterprise OID extension — the same general technique (jurisdiction-specific identity attributes carried in a custom OID inside an X.509-family certificate) that China's own GM/T 0015 SM2 certificate standard uses for its Unified Social Credit Identifier. That structural parallel raised two further questions: would the 2026-09-29 White House accord be compatible with how China actually regulates AI, and could c=US's own architecture, as documented, plausibly support genuine shared US-China agentic control rather than a US-only one?
Scope note. This is a technical and regulatory-structure comparison, not a claim about any government's legitimacy, intentions, or human-rights record, and not legal or diplomatic advice. It compares specific, sourced requirements and public statements to specific, sourced designs — nothing here should be read as taking a position on the merits of China's or the United States' AI policy.
Background · why “c=US”
The directory naming this project borrows has its own US national-security lineage
Not a Jev judgment — a historical fact check, kept separate from the scored analysis above and below.
Verified by search. RFC 1943, “Building an X.500 Directory Service in the US” (May 1996) was authored by Barbara Jennings of Sandia National Laboratories, and uses this worked example for its own directory entry: c=US, o=Sandia National Laboratory, ou=Employees, cn=Barbara Jennings — the same c=US country-level root this project's name plays on. The same RFC documents InterNIC as an active X.500 directory-service provider of that era (holding data for 52 organizations by mid-1995) and notes DOE-funded ESnet's role in the surrounding US research-network infrastructure.
As recounted by the project's author, not independently re-confirmed in full. The student reports that Sandia was also instrumental in pushing X.509v3 certificate support into this same US X.500 effort, working alongside ds.internic.net, in service of Sandia's Department of Energy nuclear-weapons mission need for strong device and personnel authentication. This document search confirmed Sandia's authorship of the US X.500 effort and its c=US example entry, but could not independently confirm the specific X.509v3 detail, the ds.internic.net hostname, or the nuclear-weapons framing from public sources — it is recorded here as the student's own contributed context, not a Jev-scored or independently re-verified claim.
Method
Same split as the accord scorecard: two different jobs
Every fact and framing choice below came from Claude Code, verified by search against primary and reported sources. Every numeric judgment came from Jev, given that fixed state.
- Given those fixed, sourced descriptions, returned every score, confidence, and probability distribution shown below for how compatible the accord is with each Chinese requirement, and how feasible each factor makes genuine shared US-China agentic control under c=US's current design.
- Jev did not choose the requirements, fetch any source, or decide what counts as a "blocker" versus a "moderate extension" — it only judged the pre-written rubric against the pre-written state.
- Full request/response JSON, including model version and token usage, is in Sources.
Result 1 of 2
Is the White House accord compatible with Chinese AI regulation?
Scored 0 (not compatible) to 3 (substantially compatible) against five specific, verified Chinese requirements and preconditions.
Every dimension scored at or below 0.14 out of 3, each with 86–94% confidence — not close, and not close with unusual certainty for a Jev run.
OVERALL VERDICT
Asked directly whether the accord, as a general model, would be recognized by Chinese regulators as a credible substitute for their own requirements, Jev returned 0 out of 2, confidence 1.00 — unanimous probability on "fundamentally incompatible."
100% · fundamentally incompatible
Result 2 of 2
Could c=US, as designed today, support genuine shared US-China control?
Scored 0 (not feasible as-is) to 3 (already feasible) against five factors spanning cryptography, governance, law, politics, and existing diplomacy.
Political/definitional alignment is the clearest blocker (76% confidence it's "not feasible as-is"). The existing bilateral dialogue channel scores highest — not because c=US uses it, but because a real, if fragile, venue already exists rather than needing to be invented.
OVERALL VERDICT
Asked directly how close c=US is today to a workable basis for genuine shared control, Jev returned 0.47 out of 2, confidence 0.29 — the least confident verdict in this project's Jev comparisons so far.
53% · not close
47% · useful reference point, not a solution
Zero probability landed on "a workable starting basis" — Jev would not call c=US, as documented, either a dead end or a viable starting point for joint control.
What this checked, and what it found
The accord and Chinese AI regulation are structurally opposite
Voluntary, self-selected, no-government-inspection versus mandatory filing, government approval, and ongoing state audit. This isn't a values judgment about either system — it's that neither could presently stand in for the other.
c=US's OID technique has a real Chinese structural parallel — not a shortcut to interoperability
Both c=US's demonstration certificates and China's GM/T 0015 SM2 standard embed jurisdiction-specific identity attributes via custom OID extensions in an X.509-family certificate. That's a shared pattern, not shared cryptography, registry, or trust root — the algorithms and the OID arcs themselves are unrelated and not currently bridgeable without new work.
Politics, not protocol, is the tallest blocker to shared control
Every technical factor (crypto mismatch, root federation, data localization) scored in the "feasible only with a major redesign" range. The political/definitional factor scored worse and with higher confidence — the disagreement over who gets to define "AI safety" is harder than any of the engineering.
A diplomatic channel already exists, even if c=US doesn't use it
The Geneva 2024 and Beijing 2026 dialogues mean a shared control-plane proposal wouldn't need an invented venue — only a jointly-governed redesign, which c=US does not currently describe.
Sources and underlying data
What this page draws on
- China's Interim Measures for the Management of Generative AI Services (2023) and the 2022 Algorithm-Recommendation Provisions, as reported by China Briefing, the Future of Privacy Forum, and Hogan Lovells.
- China's Personal Information Protection Law (PIPL) Article 38 and the Cybersecurity/Data Security Law data-localization regime.
- China's 2026-08-31 stated preconditions for AI-safety talks and the World AI Cooperation Organization launch: TechTimes, PBS NewsHour, Caixin Global, The Diplomat, and Al Jazeera, all dated 2026.
- The existing US-China AI dialogue track and the reported agent-risk agenda for proposed September 2026 talks: Brookings, TechBooky, and Sandia National Laboratories' “Challenges and Opportunities for US-China Collaboration on Artificial Intelligence Governance” (2025) — which itself frames China's stated position as wanting AI made verifiable, regulatable, traceable, predictable, and trustworthy, and calls for international AI governance frameworks that respect different countries' principles and practices, the same tension this page's political/definitional-alignment score reflects.
- Sandia's own historical role in the US X.500 directory effort this project's
c=USname plays on: RFC 1943, “Building an X.500 Directory Service in the US” (Barbara Jennings, Sandia National Laboratories, May 1996). - China's GM/T 0015 SM2 digital-certificate-format standard and the Unified Social Credit Identifier (GB 32100-2015), via the GM-Standards project and Wikipedia; c=US's own certificate profile:
research/openssl with oid.txtandschema/cus-registry.schemain this repository. - Raw Jev request/response data for every number on this page:
research/jev-accord-china-regulation-compatibility.jsonandresearch/jev-cus-china-us-shared-control-feasibility.jsonin this repository. - See also the White House accord scorecard this page follows from.