A register of unresolved questions in AI policy that current sources raise but do not answer: contradictions between high-confidence claims, and pending decisions whose outcomes would reshape the surrounding coverage. Each entry pairs a question with the sources pulling on it and a resolution condition stating what observable event would settle it.
Active questions
Q1 — Scope of the Trump administration's AI executive order(s)
Whether the Trump administration ships one or two AI executive orders, and which would govern, is unresolved. Reporting describes two distinct possible tracks: a broad model-review EO characterized as "FDA-like" by Kevin Hassett (Politico, May 5), and a narrower cybersecurity-only EO that omits mandatory pre-release testing (Bloomberg, May 8). The May 5 CAISI agreements with Google, Microsoft, and xAI bear on which approach prevails. The outcome determines whether AI Pre-Release Vetting becomes a procurement gate, a soft-law compliance gesture, or a binding regulatory regime, and conditions the trajectory of Procurement-Driven AI Governance and the standing of state-level alternatives such as California SB 53 and Colorado SB 26-189.
Resolution condition: EO(s) actually published in the Federal Register, or an explicit administration statement abandoning one track. Watch the next 90 days.
Promoted from Wiki/_meta/briefings/2026-05-10-brief.md ("One question worth sitting with") and Wiki/_meta/briefings/weekly-2026-W19.md (thesis section).
Q2 — Recursive AI R&D timeline versus the governance clock
Jack Clark's Import AI 455 estimates a 60%+ probability of recursive AI R&D by end-2028. If that estimate holds, whether any procurement-driven or pre-release-vetting governance regime could be in place in time is unresolved. Resolving it bears on whether governance coverage anchored on AI Pre-Release Vetting, AI Safety Frameworks Compared: RSP, Preparedness, NIST RMF, and Safety Cases, and the the standards/ pages family remains the right frame, or whether attention should shift toward post-recursive-improvement governance modes such as interpretability, autonomous-agent deployment liability, and model-supply-chain attestation. Related forecasting context appears in Situational Awareness: The Decade Ahead and AGI Timelines.
Resolution condition: This is a forecasting question, not a fact question. Resolution = (1) the end-2028 milestone passes (~32 months), or (2) an empirical observation falsifies the trajectory, e.g., a major frontier lab publishes a capability plateau result against the benchmarks Clark cited. Track via Track Record.
Promoted from Wiki/_meta/briefings/weekly-2026-W19.md (thesis and contradictions sections).
Q3 — Whether the IMF Mythos designation moves policy
On May 7, the IMF designated the Claude Mythos preview and GPT-5.5-Cyber as macro-financial-risk vectors, the first IMF designation of specific frontier models as systemic-financial-risk vectors (IMF financial-stability blog, May 7, 2026 — to be ingested as foundational). Whether this prompts central banks (Fed, ECB, BoE), national bank supervisors, or stress-test frameworks to add AI-cyber as a category, or remains a blog-post category change without operational follow-through, is unresolved. If it propagates, it would open a macro-prudential AI policy track not yet developed in the coverage; if not, it stands as a single-source novelty.
Resolution condition: A central bank or national supervisor (not a researcher in a personal capacity) issues guidance, a stress-test scenario, or a mandatory disclosure that explicitly references AI capability as a financial-stability category. Watch 6–12 months.
Promoted from Wiki/_meta/briefings/2026-05-10-brief.md and Wiki/_meta/briefings/weekly-2026-W19.md (thesis section).
Q4 — Anthropic's commercial standing versus political exclusion
In the same week Anthropic raised $50B at a $900B valuation (FT, May 8), it was excluded from the Pentagon classified-network cohort (xAI, OpenAI, Google, Nvidia, Reflection, Microsoft, and AWS; coverage May 1). Whether this divergence is a temporary political artifact (reported Anthropic-Sacks tension and Anthropic-MAGA tension over Mythos messaging) or a durable structural split between commercial-frontier success and political-frontier influence is unresolved. The answer bears on Musk v. Altman (and OpenAI / Microsoft / Brockman) outcomes, alignment of AI Pre-Release Vetting, and the procurement-driven-governance dynamic. Related developments include the Anthropic-SpaceX Colossus 1 deal (May 6), the Anthropic-Akamai $1.8B deal (May 8), and WSJ reporting (May 8) on a Mythos→Cairncross→EO drafting chain.
Resolution condition: (1) Anthropic added to a subsequent classified-network cohort tier within 12 months, or (2) Anthropic's commercial momentum reverses (a quarter of negative ARR growth or a missed major deal).
Promoted from Wiki/_meta/briefings/weekly-2026-W19.md (thesis section).
Q5 — Federal versus state preemption on AI
With xAI v. Colorado (DOJ intervening on 14th Amendment grounds, May 6) and the parallel American Leadership in AI Act (Lieu/Obernolte, May 6), the federal preemption fight over state AI regulation has gone live. Whether it produces an actual preemption ruling, an EO, or a federal-floor / state-ceiling compromise is unresolved. The outcome bears on Colorado AI Act (SB 24-205), California SB 53, the Idaho, Nebraska, Tennessee, and Utah chatbot laws, and Pennsylvania v. Character.AI: a preemption ruling would moot much of the state-law track, while a federal floor would amplify it. Related items include the GUARD Act Senate Judiciary advance (May 6) and Connecticut SB 5 (May 6).
Resolution condition: Court ruling on xAI v. Colorado, or enacted federal AI legislation that explicitly addresses state-law preemption.
Promoted from State-Level AI Regulation, flagged via the May 4 ingest cycle.
Resolved (last 90 days)
(Empty — populated as questions are resolved. Each resolved entry stays here with the resolution date and a one-line outcome before being moved to the archive on the 90-day anniversary.)
Archive
(Older resolved questions migrate to Wiki/analysis/open-questions-archive-YYYY-Q.md files. None yet — this page was created on 2026-05-10.)
How to add a question
- Via daily brief: the
daily-briefskill's "One question worth sitting with" promotes here if it survives 7 days. - Via weekly synthesis: the
weekly-synthesisskill's "Three highest-leverage threads" can spawn questions if framed as unresolved. - Via query workflow: when a substantive
queryinvocation reveals a knowledge gap, thequeryskill appends to "Recently asked". - Via direct add: drop
OPEN-<topic>.mdinWiki/queue/. Thequeue-processorskill formats it and inserts here.
Recently asked (queue from query workflow)
(Questions surfaced by the query skill that haven't been formally promoted to Active yet. Each one waits here until the next weekly-synthesis pass evaluates whether it deserves promotion.)