Three leads showed up after the main pack was already standing. Steve Jurvetson posted charts saying foreign accounts amplified Jacob Coxon. Rooke Poole drew a diagram that runs pause money toward lawmakers. Kevin Bass argued that METR’s funders sit on Anthropic’s cap table. Each one is a lead. A lead is not a receipt.
We checked them the same way we checked the rest of this report. If a primary names the dollar or the role, it stays. If the tip’s number and the filing disagree, we publish the filing. If the dollar is not on a primary at all, it stays off the money pages. The editorial on the DSC blog was not rewritten from these tips.
In this report: BLUF · Research map · Report assets · Amplification scoreboard · Weekend timeline · Money graph · Thesis audit · Follow-on checks · Evidence & sources
The charts are theirs until we recount them
Steve Jurvetson posted DigitalBorders charts saying Coxon’s posts were heavily amplified by foreign-located X accounts. The charts are real. They are on the post. We have not yet pulled our own sample of the reposters and tallied them. Until we do, cite them as his charts, not as our sample. The weekend timeline and the scoreboard already score them that way.
DigitalBorders is a free public research site. It studies where the people engaging with a post appear to be based, using public location signals, and splits them into US, foreign, and unresolved. There is no paid Coxon spreadsheet we failed to buy. Jurvetson’s “Source” link goes to the homepage, not a case file. The numbers live as free charts on the post.
A location percentage is not a bot network, and it is not a state. Their own method says that. We will say it too.
DigitalBorders, posted by Steve Jurvetson (@DFJSteve), September 11, 2026. Coxon’s resignation post: 3,500 reposts sampled from a reported 15,017. Foreign 2,672. United States 633. Unresolved 195. Their chart. We have not recounted the sample. File.
Same charts, same post. Thayer counter-narrative: 3,500 reposts sampled from a reported 6,124. Foreign 1,205. United States 1,987. Unresolved 308. File.
Before one of these charts goes into a DSC piece as our count, the job is: archive the graphic and the post, write down their sample size and the percentages, pull our own sample of reposts at their N when we can (they used on the order of 3,500), bucket the same three ways, and compare. Match, close, or diverges. “We couldn’t download a CSV” is not the same sentence as “the chart is fake.” The recount of the Coxon and Thayer charts is still open.
Rooke Poole’s diagram
Rooke Poole’s diagram maps capital, grantees, and campaigns from pause advocacy toward lawmakers. Useful list. We re-checked every edge against the same kind of primary as the rest of this report, on September 15, 2026. Unverified diagram dollars do not become our dollars. A matching pledge, the ‡ mark, is never added into the cash column. AIPI is not AIPN. When we show money, we show an SFF recommended figure, or another named primary.
Capital
| Edge | Score | What we publish | Primary |
|---|---|---|---|
| Musk → FLI $10M (2015) | CONFIRMED | $10M | FLI Wayback · Wired 2015-01-15 |
| Tallinn → FLI board / co-founder | CONFIRMED | role only | FLI Our people · Wikipedia: Tallinn |
| Tallinn → SFF principal client | CONFIRMED | role only | SFF FAQ (SFC primary client = Jaan Tallinn) |
| Vitalik → SFF via “Metta” | UNSUPPORTED | omit | No Metta/Vitalik→SFF primary; likely Metaplanet/MetaMed/FLI mixup |
| Vitalik major 2021 gift → SFF | PARTIAL | omit as SFF; corrected edge Vitalik→FLI | Politico (~$660–665M to FLI) |
The Musk gift and Tallinn’s FLI and SFF roles are the ones that belong on the money graph. The Vitalik-to-SFF line does not. The big 2021 gift was to FLI.
Grantee dollars, tip versus the SFF table
Primary: SFF recommendations, the SFF-2025 first row per organization. ‡ is noted and never added into the cash we publish.
| Org | Tip $ | Our SFF recommended | ‡ matching | vs tip | What we publish |
|---|---|---|---|---|---|
| Palisade Research | $1.15M | $1,133,000 | $928,000 | MATCH | $1,133,000 (+ note ‡) |
| GovAI | $750k | $756,000 | $696,000 | MATCH | $756,000 |
| AI Futures Project | $2.12M | $2,035,000 | $500,000 | MATCH | $2,035,000 (elsewhere in this pack: $1.535M cash face + $500k‡) |
| METR | $500k | $548,000 | $428,000 | DIRECTIONAL | $548,000 / ‡ as on Matching pledges |
| Encode AI Corp | $400k | $516,000 | — | DIRECTIONAL | $516,000 (already on the money graph) |
| Tarbell Center | $200k | $783,000 | $200,000 | tip = ‡ only | Publish $783,000 recommended; tip tracked ‡ — do not treat $200k as the full grant |
| CAIS | $150k | $289,000 | — | DIVERGES | $289,000 if cited |
| CAIS Action Fund | $2.12M | $772,000 | — | DIVERGES | $772,000 if cited |
| MIRI | $410k | $1,607,000 | $1,392,000 | DIVERGES | SFF primary only |
| IAPS | $7.2M | $572,000 | $300,000 | DIVERGES | SFF primary only — tip wildly high |
| AIPI (via Hack) | $3.12M | $1,635,000 | — | DIVERGES | $1,635,000 (already on the money graph) |
| CHT / CivAI / Objectives | tip high | SFF hits lower | — | DIVERGES | SFF primary only |
| PauseAI US / Global, Common Sense, Ekk8, ESP, Blueprint, YPA, Evitable, TBIJ, Oversight | tip $ | — | — | NO_HIT on SFF tables | omit $ (PauseAI protests existing ≠ an SFF receipt) |
The diagram’s “$21.5 million or more in total flows” does not hold as stated. SFF’s own 2025 year card is about $35 million. A re-sum of the first rows, matching pledges excluded, is about $34.3 million. The diagram’s “SFF 2024 $24.72 million” is close to a re-sum of the 2024 recommendations plus FlexHEGs, $23,955,756, and not the same number.
Campaigns and bills
| Node | Score | How to cite |
|---|---|---|
| FLI Digital Media Accelerator | CONFIRMED | FLI project page |
| Protect What’s Human | CONFIRMED | protectwhatshuman.org (linked from FLI) |
| Facing the Frontier | UNSUPPORTED | omit as an FLI campaign node |
| PauseAI protests | CONFIRMED activity | pauseai.info — not an SFF $ edge |
| UK “ASI Security Bill” | PARTIAL | Official title Artificial Superintelligence Bill — bills.parliament.uk/bills/4288 (Sobel, First Reading 2026-09-08). Not law. |
| UK AI kill-switch amendment | CONFIRMED | Lords amendment to Cyber Security and Resilience Bill (BBC). Not law. |
| US superintelligence-ban effort | CONFIRMED announced | Sanders/Casar Ban Artificial Superintelligence Act — Senate press 2026-09-03 |
We are not claiming any diagram dollar that lacks a primary, especially the PauseAI-neighborhood tips and the IAPS $7.2 million. We are not claiming Vitalik funded SFF, or a “Metta” vehicle. “Facing the Frontier” is not an FLI program on the pages we checked. The UK and US bills are efforts. They are not law.
Diagram: Rooke Poole, status/2099660837388501127.
| What | Live | Receipt |
|---|---|---|
| Musk → FLI $10M (2015) | FLI Wayback | Wayback HTML |
| SFF recommended tables | SFF recommendations | HTML · 2025 |
Kevin Bass’s thread
Kevin Bass’s thread argues that METR’s funders overlap Anthropic equity holders, that a roughly $71 million raise landed beside talk of an evaluator seat, that Tarbell journalism sits on Coefficient and Open Philanthropy money, and that Redwood and Coefficient ties, including Barnes and Greenblatt, close a loop through Moskovitz philanthropy. Useful list. We re-checked the edges against live primaries on September 14–15, 2026.
Three corrections up front, because the thread’s round numbers travel faster than the filings. $7.7 billion is ceiling math, not a stamped Good Ventures holding in Anthropic. Tarbell’s Coefficient and Open Philanthropy class is $4.7 million, not $5.3 million. Redwood’s subcontract on the public METR record is the OpenAI and Hugging Face probe, not Anthropic. “The money anticipated the seat” stays a read. It is not a fact. This tip did not rewrite the DSC editorial.
Kevin Bass, @kevinnbass, thread. His graphic. The $7.7 billion mark on it is the ceiling above, not a holding we stamp. File.
Moskovitz, the Anthropic stake, and the philanthropy
| Edge | Score | What we publish | Primary |
|---|---|---|---|
| Early-2025 donated Anthropic stake into a nonprofit; Forbes ~$500M / True Net Worth <0.8% | CONFIRMED | donation + estimate band; vehicle unnamed | Forbes AU · Forbes True Net Worth 2026-04-20 |
| Ceiling math 0.8% × $965B Series H ≈ $7.7B | PARTIAL | ceiling only — never as stamped book value | Anthropic Series H |
| Moskovitz: “We fund … METR and Redwood”; Good Ventures “beneficiary … via Anthropic” | CONFIRMED (words) | quotes + grant-path hedge | Bluesky 2026-09-11 · Bluesky 2026-08-26 |
| Cash path OP/GV → ARC → METR; Coefficient/OP → Redwood | PARTIAL (loop) | cite the money graph — not direct OP→METR; not payroll | same |
Who METR will take money from
| Edge | Score | What we publish | Primary |
|---|---|---|---|
| METR refuses frontier labs + employees (and donations at their direction); does not name lab investors as barred donors | CONFIRMED | policy text as written | metr.org/about · Funding update 2026-08-14 · coi-policy.pdf |
| Tallinn / SFF → METR/ARC path and Anthropic Series A lead | CONFIRMED | dual path; “observer” only on secondary sources | Anthropic Series A · SFF recommendations |
| Schmidt Sciences → METR and Eric Schmidt on Series A | CONFIRMED | person vs vehicle split | metr.org/about · Anthropic Series A |
| Jane Street individuals → METR; Jane Street firm → Anthropic equity | PARTIAL | both edges true; not the same legal person | metr.org/about · Anthropic Series E · Series F |
| McClave / BEMC Anthropic investor | CONFIRMED equity; no METR grant found on public 990s checked | equity yes; METR donor no — Bass already softens this row | Anthropic Series A/B · CNN 2024-03-10 |
The donor ban names labs and their employees. It does not name the people who own the labs. That silence is the conflict. It is not a payroll.
Kevin Bass, same thread. His table of the overlap. The ban we checked is the paragraph above: labs and employees, not investors. File.
The $71 million
Kevin Bass, same thread. The $71 million and the August 14, 2026 announcement are METR’s own update. The sentence on his graphic that the money came before the seat is his reading. We did not adopt it. File.
| Edge | Score | What we publish | Primary |
|---|---|---|---|
| METR ~$71M commitments in prior 6 months, announced 2026-08-14 | CONFIRMED | $71M + window | Funding update |
| Calendar overlap with independent-evaluator / FINRA-for-AI drafting | CONFIRMED (overlap) | timing only | Funding update window vs policy primaries in the Bass timeline (OpenAI Frontier Safety Blueprint, Anthropic AI Exponential policy, Hassabis framework, Bloomberg FINRA-like watchdog, H.R.9925, Lawfare FINRA-for-AI) |
| OpenAI HF investigation agreed ~Jul 29–30; Anthropic METR agreement Sep 9 (end/after raise) | CONFIRMED | sequence | METR HF report · Anthropic alignment assessment |
| Money anticipated a mandated evaluator seat | UNSUPPORTED as proven | omit as fact — Bass’s own hedge stands | same chronology |
| 7× 2024 budget | PARTIAL | omit until a public budget primary exists | — |
Drafts that name METR
| Edge | Score | What we publish | Primary |
|---|---|---|---|
| Sample Apr–Sep 2026 docs: Amodei / Sacks / Lawfare-class name METR; most institutional drafts silent | PARTIAL | directional pattern; not a hard “exactly 4 of 22” stamp without one still-unverified named cell | Amodei / Sacks / Lawfare sample vs institutional silence |
| Amodei Pace the Frontier names METR + permanent embeds | CONFIRMED | already scored on the amplification scoreboard | darioamodei.com |
Tarbell
| Edge | Score | What we publish | Primary |
|---|---|---|---|
| Coefficient/OP-adjacent awards tied to Tarbell = $4,703,000 (3 archive rows) | CONFIRMED | $4.7M — not Bass’s ~$5.3M | Coefficient Giving Grants Archive CSV (Training for Good $999k + Tarbell $816k + Tarbell $2,888k) |
| SFF → Tarbell ~$1.3M incl. $200k‡ | CONFIRMED | recommended cash + ‡ face kept separate | SFF recommendations · Rooke table above |
| Fellows place AI pieces at TIME / Verge / Dispatch / LAT / Science | CONFIRMED | placements | tarbellcenter.org |
| Consistent in-story funding disclosure | PARTIAL | author-bio common; in-article uneven | same |
Redwood, Coefficient, Barnes, Greenblatt
Kevin Bass, same thread. The board lines and the Coefficient recommendation match the filings in the table. The graphic’s claim that Redwood subcontracted METR’s Anthropic investigation does not. The public record is OpenAI and Hugging Face. File.
| Edge | Score | What we publish | Primary |
|---|---|---|---|
| Redwood board eras include Karnofsky / Cotra / Christiano lines | CONFIRMED | roles | Form 990 EIN 87-1702255 (ProPublica Nonprofit Explorer) |
| Coefficient >$70M over two years recommendation to Redwood (Sep 2026) | CONFIRMED | >$70M recommendation | Coefficient: urgently scaling |
| Redwood staff contracted to METR on the OpenAI / Hugging Face investigation (Ryan Greenblatt) | CONFIRMED | OpenAI/HF only | METR HF investigation |
| Same fact sold as “Redwood subcontracted on the Anthropic investigation” | MISATTRIBUTED | omit the Anthropic-subcontract frame | Anthropic’s Sep 9 alignment assessment names METR only |
| Greenblatt = domestic partner of Beth Barnes (METR CEO); Barnes recused from the engagement decision | CONFIRMED | disclosure footnote | METR HF report, footnote 2 |
The loop, as far as it goes
The rows above are a setup. They are not an invoice.
| Claim | Score | How we cite |
|---|---|---|
| Equity → philanthropy → evaluator/grantee incentive loop (Moskovitz/Tallinn neighborhood + METR donor-ban silence + $71M timing + Redwood dual-hat) | CONFIRMED as setup. The apparatus is visible. The stamp is not. | This page + the pacing row on the scoreboard |
| METR is “on Anthropic’s payroll” / literal capture invoice | UNSUPPORTED | omit |
| Network “cannot turn off doom cycle without killing funding” | PARTIAL | hypothesis, not a mechanism we can show |
| Coxon weekend = product of that machine | adjacency only | do not reopen a Coxon payroll claim; cite the activation rows on the scoreboard |
We are not claiming Bass’s ~$5.3 million Tarbell figure. We are not claiming $7.7 billion as money Good Ventures holds in Anthropic today. We are not claiming METR is on Anthropic’s payroll. We are not claiming Redwood subcontracted the Anthropic investigation. We are not claiming the $71 million was raised because a seat was coming. We are not claiming a hard count of “exactly 4 of 22” policy drafts. We are not rewriting the editorial from this thread.
Thread: Kevin Bass, @kevinnbass, status/2099621874279817638.
| What | Live | Receipt |
|---|---|---|
| METR ~$71M update | funding update | HTML |
| METR about + COI | about · coi-policy.pdf | about HTML · PDF · txt |
| OpenAI / HF investigation | metr.org | HTML |
| Coefficient archive → Tarbell $4.7M | CSV | full CSV · Tarbell slice |
| Coefficient >$70M Redwood | urgently scaling | HTML |
| Anthropic Series A / Series H | Series A · Series H | A HTML · H HTML |
| Amodei Pace the Frontier | darioamodei.com | HTML |
| Anthropic Sep 9 METR agreement | alignment assessment | HTML |
| Bass tip figures | thread | tips/ |
| Jurvetson / DigitalBorders charts | as published on X | chart1 · chart2 |
Full drawer: Evidence & sources