9 September 2026 local campaign date. User requested a look at Coxon's viral resignation before proceeding. Fresh check confined to public research credentials, the statement, directly relevant funding/advocacy links and chronology. No outreach, graph edit, ZIP rebuild or visible UI. r33 contains no Coxon or hilbertspaess record; these findings remain a separate supplement.
Assessment
Coxon has a verifiable technical publication record. OpenAI's GPT-4o system card names him among contributors. The six-author paper Weight-sparse transformers have interpretable circuits credits him with optimization and circuit-pruning work, tasks and a circuit visualizer. Reading scope: author list and contribution statement, not full technical replication or verification of all employment dates.
[GPT-4o system card](http[local research file] PDF26; [sparse-transformer paper](http[local research file] PDF1 and32. OpenAI's emergent-misalignment paper thanks both Coxon and Evan Hubinger for feedback; shared acknowledgment does not itself establish direct collaboration between them.
His seven-post statement reports resignation, attributes private fears to lab colleagues, forecasts dangerous capabilities and advocates coordinated pacing with a possible temporary capability-development ban. These are distinct testimony, forecasts and policy positions. The thread supplies no underlying internal documents or quantitative extinction-risk derivation. Native X failed; all seven posts were read through the [Thread Reader mirror](http[local research file] September8 US evening versus September9 UTC is consistent with the displayed date difference. Exact native rendering was not inspected.
Financial record
The ProPublica-rendered IRS return for Good Ventures Foundation, EIN46-1008520, TAX YEAR2022, covering1July2022 to30June2023 and submitted15May2024, names JACOB COXON in PartXV3a Paid during the year. Amount USD20,159; purpose Long-Term Future Scholarship Program - Research Scholarship. Exact-name scholarship recipient match is verified in the filing; independent confirmation that this is the researcher has not been acquired. The earlier fiscal-year-endingJune2022 return had no Coxon text match. It is inaccurate to silently treat the TY2022 label as a proved calendar2022 payment date.
The same return names Cari Tuna as president/director and Dustin Moskovitz as secretary/treasurer/director. This connects the named scholarship to the foundation's governance. It supplies no present campaign compensation, scholarship conditions, continuing obligation or instruction concerning this resignation. Personal recipient address is deliberately omitted from this note; the public filing remains in the separate original capture.
[Full filing landing page](http[local research file] [rendered IRS990PF](http[local research file] Ordinary public GET supplied the readable return after the web reader could not fetch the embedded rendering. Scoped reading: header/tax period, named grant row and paid-section label, officer rows; not full financial audit.
Advocacy and institutional response
Andrea Miotti's9September ControlAI article leads with Coxon's resignation, then describes ControlAI's UK/US legislative work. This establishes use of the statement in advocacy. Miotti claims ControlAI supplied the UK bill, consulted on the Sanders-Casar framework and drafted a government kill-switch amendment. These role claims are attributed to ControlAI. The article distinguishes its narrower precursor/ASI approach from the US proposal's broader temporary advanced-AI pause. It establishes neither sponsorship of Coxon's departure nor advance media coordination.
[Miotti article](http[local research file] Full substantive article read via ordinary publisher GET. Government chronology checked separately: Sanders' official3September announcement describes forthcoming legislation, permanent ASI prohibition, a temporary advanced-development pause and a new cabinet-level regulator/advisory board. That announcement predates Coxon's resignation. Announcement is not enactment or a complete bill-text analysis.
[Sanders announcement](http[local research file]
ABC reports Hubinger publicly agreed with the danger concerns while saying Anthropic was trying its best and lacked a settled superintelligence-alignment plan. His personal greater-than10percent next-decade estimate is not an established probability. Native X reply was not fetched; [ABC's report](http[local research file] supplies the accessible attributed text. Existing publication overlap is a separate and weaker relationship than this response.
Coxon told Axios he left before Anthropic equity vested, retains OpenAI equity, and had not observed Anthropic compromising safety. [Axios interview](http[local research file]
Forest and next discriminator
The supported structural finding is public translation of a technically credentialed researcher's warning into an existing legislative campaign. Scholarship history, sincere conviction, publication relationships, media planning and advocacy use can coexist, but are distinct propositions. No acquired evidence establishes paid resignation, covert direction, an NGO employment role or private coordination of the rollout. Claims about first quote-post order and precise amplification timing encountered in social commentary were not verified and are not promoted.
Most informative next records, if pursued: recipient confirmation and scholarship terms; actual advance briefing/communications arrangements; Coxon's next institutional role or funding; and whether policymakers explicitly use his testimony to justify particular operative powers. Broad acquisition stops at this bounded check. The existing consolidated explorer remains frozen.