# Six posts for the "cabal" section, verbatim, with every claim graded Pulled 24 September 2026, 11:30 UTC, through `api.fxtwitter.com//status/` (a `User-Agent` header is required; without it the API returns 404). Counts are as of the pull. Images were downloaded from `pbs.twimg.com` and read; the descriptions below are of the image, not the post text. Quoted posts were pulled the same way. Threads: the first Bass post promises "Evidence and Github in next post"; fxtwitter does not return thread replies, so the evidence was taken from the GitHub repository the figure itself cites (`github.com/kevinnbass/metr-money-figure`, v0.45, 14 Sep 2026), cloned and read. Part A is the six posts and their quoted posts, verbatim. Part B is every factual claim about money, people and institutions, one row each, with a grade and the source it was checked against. Grades: - **VERIFIED**: a primary source was read (an IRS filing, a company announcement, a grant-maker's own page, the person's own words in a transcript). - **REPORTED**: only credible press says it (named outlet, named reporter), or the primary source is paywalled and a capture could not be read. - **CLAIMED**: only the post says it, or it is the poster's own count or inference. --- ## Part A. The posts ### 1. Kevin Bass, "I have conducted an audit of Anthropic's finances" - Author: Kevin Bass (@kevinnbass), 234,299 followers - Date: Mon 14 Sep 2026, 22:10:38 UTC - Likes 34,066 · reposts 9,647 · quotes 2,049 · replies 1,555 · bookmarks 22,678 · views 6,525,963 - Link: https://x.com/kevinnbass/status/2099621874279817638 - Format: long-form post (X "note"). Not an X Article. Text, verbatim: > I have conducted an audit of Anthropic's finances. > > What I have found is so shocking that I am calling for a Congressional investigation. > > Anthropic is not just seeking regulatory capture. > > It has built a regulatory capture machine that cannot be turned off. > > Structural financial incentives make it impossible for Anthropic -- I call it the Anthropic Network -- to turn off its own AI doom cycle. > > It starts with METR. > > Dario Amodei proposes "third-party evaluators" to assess the risk of Anthropic's models. > > He proposes METR for this purpose. > > But METR is financially dependent on the Anthropic's success -- specifically, on the explosive growth of more than $7 billion dollars in Anthropic stock. > > Dustin Moskovitz invested this stock into Good Ventures Foundation, where it represents the majority of that organization's portfolio. > > And GVF is the overwhelming funder of the entire Anthropic Network ecosystem. > > This stock was worth $500 million early last year. > > It is worth more than $7.7 billion just ~16 months later. > > METR -- and all of those building a career its parent organizations -- cannot afford to disrupt that growth. > > Because if Anthropic goes under, many of the organizations that fund METR go under as well. > > But if Anthropic succeeds, METR and its parent organizations become more richly financed to regulate AI -- something those at METR want very much. > > The "third-party evaluator" is not "third-party" at all. > > The evaluator is on Anthropic's payroll. > > If this were the end of it, that's bad. > > But that isn't all. > > The same organizations that fund METR also fund the many organizations, such as the Tarbell Center, that promote AI Doom. > > The Tarbell Center publishes AI Doom articles in The Verge, Science, LA Times, The Dispatch, TIME, and others. > > They are selling the problem, and then selling the solution to the problem -- from the same money pile: Anthropic's. > > All of these organizations are financially dependent on the same exploding $7 billion money pile. > > As Anthropic grows more and more powerful, its AI Doom Machine grows better and better financed -- louder and louder. > > Meanwhile, the regulatory regime seeded in METR grows larger to solve the increasingly loud -- now hysterical -- problem of AI Doom that the Anthropic Network itself created. > > From this standpoint, as Anthropic becomes more powerful, AI might be getting scarier, sure -- but the positive feedback loop also becomes more deafening -- independent of objective facts. > > This itself is an objective fact. > > The deafening AI Doom is part of an business model, that, as it expands, so too does the AI Doom messaging -- there is simply more money to do it. > > But the problem also goes in the other direction: > > If Anthropic dies, the Regulatory Regime and the AI Doom Machine are crippled or die. > > Neither METR nor Tarbell nor the other organizations in the Anthropic Network can allow that to happen. > > Hence, neither METR or the AI Doom Machine can be trusted to provide independent assessments of Anthropic's models or AI more broadly. > > They simply are not organizations independent of Anthropic. > > And Anthropic cannot detach itself from METR or Tarbell or countless other safety orgs (not shown here), either, because they drive hype for the models and the possibility of eventual regulatory capture, and Anthropic will not give that up willingly. > > What's more, the people at all of these organizations are all the same ecosystem, the same community. They just shuffle between organizations. > > The Anthropic Network is therefore, so long as it is successful, locked into a self-amplifying feedback loop inside an ideological monoculture. > > And that feedback loop is winning. > > That's what Jacob Coxon is. > > China is keeping messaging tight. That is why optimism for AI is so high in China. > > America has Anthropic: a massive company pushing anti-AI propaganda at a state level. > > Anthropic will either create hysteria until American AI slows down and China wins, or it will create fractures throughout American society with severe political consequences. > > Ironically, because of the structural financial incentives underpinning the Anthropic Network, it has become the same kind of self-amplifying virus that it fantasizes AI to become in the future -- while hiding its tracks just as carefully. > > It is the mirror of the same AI virus that it hypothesizes to consume America. > > Anthropic's business model, models itself after the very thing it claims to fear. > > Except Anthropic's ideology infects humans, not computers. > > Congress must investigate. > > Evidence and Github in next post. > > Then some supplementary figures. Image (one, 3520×4096, https://pbs.twimg.com/media/HSNXaLrW4AEnwuB.jpg): a dense money-flow (Sankey-style) chart in blue, orange, yellow and grey on cream, titled "FUNDERS → INTERMEDIARIES → THE EVALUATOR, AND THE LAB → THE FUNDERS". Headline, verbatim: "Funders that reach METR through ARC, RAND, Longview and pooled funds include Anthropic's Series A investors, two of them its board observers by their own account; Moskovitz donated a stake now worth up to $7.7B to what he calls 'our foundation', and no filing checked shows where it sits; Good Ventures has no METR grant on its books". Left column: Anthropic (Series A May 2021; Series H $965B), Dustin Moskovitz (Coefficient Giving, Good Ventures Foundation, "$10.1B endowment (Jun 2025)"), Jaan Tallinn / SFF ("Anthropic Series A lead, Series B; board observer by his own account"), The Audacious Project, donor-advised funds. Middle column: Alignment Research Center ("METR's parent until Dec 2023; handed METR $4.55M at the spin-out"), RAND Corporation ("METR's partner in Canary"), Longview Philanthropy, FAR AI, Redwood Research ("Worked with METR on the OpenAI investigation; Redwood says its staff are subcontracted by METR"), Constellation ("The Berkeley office METR works from"), Tarbell Center for AI Journalism (with outlet chips: TIME 45, The Verge 37, MIT Tech Review 26, Lawfare 22, The Guardian 26, LA Times 14). Right: METR. Labelled pipes include: "$1.5M 2022 → ARC"; "$10.0M 2025 'AI Evaluation and Testing' → RAND"; "$26.3M 2022–25 → Longview"; "$59.3M 2021–25 → FAR AI"; "$5.6M 2022–24 SFF recs → ARC / ARC Evals"; "$220K 2023 (per GWWC) → METR pooled fund, undisclosed"; "$752K SFF recs 2024–25 → METR incl. conditional match"; "$4.2M → METR 2024–25: Vanguard $4.0M (donor unnamed); FP $184K + SVCF $20K match Tallinn's ledger"; "$38.0M Audacious commitment to Canary (RAND + METR)"; "~$17M committed to METR"; "~$4.6M spin-out transfer, Apr 2024 (ARC 990)"; "$70M+ 2026 recommendation → Redwood"; "$2.4M SFF recs 2022–23 → Redwood"; "$10.8M 2022–25 → ARC"; "$22.9M 2023–24 → Constellation"; "$10.0M 2024 SVCF → Constellation"; "$5.3M Coefficient awards 2024–25 → Tarbell Center". A red line reads "Direct: none found. No METR-named grant in Coefficient's 2,911-row index (Sep 11 2026) or Good Ventures' 990-PFs to Jun 2025". Bottom-left bar chart, "Two published bounds on the donated stake, at different dates": "$500M early 2025 transfer (Forbes estimate, published Nov 2025)" and "≤ $7.7B Sep 2026 ceiling (0.8% × $965B Series H post-money)". Three quote boxes: Forbes Nov 2025 ("Their Anthropic stake (worth an estimated $500 million) was moved into a nonprofit vehicle in early 2025..."); Moskovitz on Bluesky 30 Mar and 11 Apr 2026 ("Our Anthropic shares are entirely in our foundation - no personal benefit"; "The foundation is invested in Anthropic as well"); Moskovitz to Stratechery 20 Oct 2025 ("I'm a chair of at Open Philanthropy and I'm a business guy and I'm also in the boardroom in at Anthropic"). Footer cites `github.com/kevinnbass/metr-money-figure` and says data collection was done with Claude Code and audited with Codex and Grok agents. ### 2. Kevin Bass, "Behind the viral Coxon story" - Author: Kevin Bass (@kevinnbass) - Date: Sat 12 Sep 2026, 01:28:44 UTC - Likes 1,020 · reposts 316 · quotes 41 · replies 44 · bookmarks 301 · views 98,692 - Link: https://x.com/kevinnbass/status/2098584562620047789 - Format: long-form post. Text, verbatim: > Behind the viral Coxon story is a tightly financially entangled regulatory capture cartel. The story is almost too clean to beelieved. > > Almost everyone involved in the past couple of weeks leading up to the viral Coxon story are paid for by investors in Anthropic. > > Good Ventures and Open Philanthropy/Coefficient Giving is owned by Dustin Moskowitz, who led Anthropic's Series A. > > Survival and Flourishing Fund funded others who pumped up the hysteria in parallel. Survival and Flourishing Fund are run by Jaan Tallinn -- an investor in Anthropic. > > Then you have amplifiers of Coxon, who are at Anthropic itself. > > All of these individuals serve to benefit financially from the doomer hysteria and the regulatory capture that could result in its wake. > > Coxon himself is a nepo baby -- he doesn't need the money. He just harvested the virtue signal. > > Hence the smug smile while talking to interviewers about humanity's impending extermination. > > Nepos gonna nepo. Images (two). First (4096×2607, https://pbs.twimg.com/media/HR-qtuVXkAA6fwt.jpg): a table titled "THE MACHINE 3 · THE PEOPLE — The roster: who touched the story, and whose money they work on. Nineteen people and organizations moved the story in its first day. Grouped by who funds them." Four groups. "Paid for by Moskovitz and Tuna (Good Ventures · Coefficient Giving)": Jacob Coxon ("The resignation, the exclusive, the thread, the group chat"; "$20,159 Long-Term Future scholarship, FY2022–23"; reach 166M views); Harry Booth (TIME interview, Sep 9; "Tarbell fellow at TIME 2024–25; Tarbell: Coefficient $6.75M"); Robert Hart (The Verge story; Senior Tarbell Fellow); Peter Wildeford ("Institute for AI Policy and Strategy: Coefficient $34.2M"); Max Nadeau (Coefficient staff); Dylan Matthews (Board of CSAIP; Coefficient program officer); Horizon fellows (staff in congressional offices; "2022 cohort: Coefficient $2.87M"). "Paid for by Jaan Tallinn (Survival and Flourishing Fund)": Sneha Revanur ("Founder, Encode: SFF $568K"); Nathan Calvin ("First quote-tweet, +6 min: 'please read'"; Encode); Daniel Kokotajlo ("AI Futures Project: SFF $2.54M"); AI Policy Institute ("SFF $2.08M"); Tarbell Center ("SFF $1.3M on top of Coefficient"). "Anthropic (Series A led by Moskovitz and Tallinn; Open Phil co-founder Karnofsky married to its president)": Evan Hubinger ("'Jacob is correct here… >10% within the next decade'"; Alignment stress-testing lead; 42M); Samuel Marks; Anna Wang; Ethan Perez ("'recruited him for ~2 years… 100% agree'"; Alignment lead); Dario Amodei / Anthropic ("Not in the thread; $1M personal and $40M corporate to Public First, 2026"). "Funder not disclosed": Data for Progress (poll fielded Sep 4–9, 68%), ControlAI, Bernie Sanders ("Bill announced Sep 3; 'Mr. Coxon is right' Sep 9"). Footer: "Money is to organizations, not to the named individuals, except the scholarship." Second image (1152×1148, https://pbs.twimg.com/media/HR-rZ1ca0AA_MVg.jpg): a still of Jacob Coxon on CNN, "LIVE", chyron "...ALL BY THE END OF THE DECADE'", smiling slightly, brown sweater, blue map background. ### 3. Parker Thayer, the scholarship - Author: Parker Thayer (@ParkerThayer), 20,059 followers; investigative researcher at the Capital Research Center (a conservative research group) - Date: Wed 9 Sep 2026, 14:26:11 UTC - Likes 224 · reposts 35 · quotes 6 · replies 7 · bookmarks 23 · views 82,990 - Link: https://x.com/ParkerThayer/status/2097693053725069548 Text, verbatim: > Worth noting: this guy got a scholarship from Dustin Moskovitz's Good Ventures Foundation in 2022. Image (1066×637, https://pbs.twimg.com/media/HRyAEBbboAAyHhk.png): a screenshot of a Form 990-PF grants schedule (Part XV, grants paid), recipients all at "314 LYTTON AVE, PALO ALTO, CA 94301" (Good Ventures' address), relationship "NONE", status "I" (individual). One row is highlighted: "JACOB COXON — LONG-TERM FUTURE SCHOLARSHIP PROGRAM - RESEARCH SCHOLARSHIP — 20,159". Neighbouring rows: Jacek Karwowski (Educational scholarship, 52,810), Jack Ryan (Century Fellowship, 134,050), James Petrie (Global Risk Research Program, 40,000), Janvi Ahuja (70,904), Jared Davis (AI Fellows Program, 140,227), Jaspreet Pannu (Technology Policy Fellowship, 61,240). Quoted post: Jacob Coxon (@hilbertspaess), Wed 9 Sep 2026, 00:04:30 UTC, second post of his resignation thread (in reply to 2097476201283834281), 132,436 likes, 15,829 reposts, 26,802,359 views, https://x.com/hilbertspaess/status/2097476203863224394: > The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible - but I hear the same people express fear privately. No other human activity poses this level of danger. ### 4. Hunter Ash, "let's meet him" - Author: Hunter Ash (@ArtemisConsort), 99,705 followers - Date: Tue 22 Sep 2026, 21:22:12 UTC - Likes 930 · reposts 89 · quotes 17 · replies 88 · bookmarks 192 · views 102,591 - Link: https://x.com/ArtemisConsort/status/2102508787915710835 - Format: long-form post. Text, verbatim: > Dustin Moskovitz is by far the single largest funder of the EA ecosystem. So let's meet him: > > Moskovitz made his fortune as a cofounder of Facebook, though he has since said he is "embarrassed to have helped it begin," citing Trump's victory, COVID misinformation, and "hate." He also co-founded Asana, a B2B SaaS company. > > He is married to Cari Tuna (pictured), a former journalist. It is unknown whether they have children. Together, they launched their nonprofit efforts in the form of Good Ventures and Open Philanthropy. The latter has since been rebranded as Coefficient Giving. > > Through these and other vehicles, Moskovitz has donated billions to causes including EA community building (Centre for Effective Altruism, Effective Ventures, 80,000 hours, and various smaller programs), criminal justice reform (leniency advocacy) including grants to BLM-related organizations, and AI safety. Coefficient says it's on track for more than a billion dollars in AI-related grantmaking in 2026. > > Moskovitz has also been a very generous political donor since 2016. He spent about $20 million in 2016 opposing Trump, $24 million in 2020 plus much more through Asana, and ~$50 million in 2024. His organizations have funded Congressional aide roles. > > He has also funded the salaries of journalists who have written anti-AI stories structured as interviews of "experts" also paid by Moskovitz orgs, as documented by @brianchau57 / Effort. > > If you want to hear from him directly, you can find him on BlueSky. Image (1000×1000, https://pbs.twimg.com/media/HS2UZzDagAAAfny.jpg): a posed portrait photograph of a couple in a kitchen, a woman in a light green sweater and cream trousers and a bearded man in a rust shirt, his arm around her; a Forbes-style portrait of Cari Tuna and Dustin Moskovitz. ### 5. Brad Neuberg, Irregular - Author: Brad Neuberg (@bradneuberg), 12,213 followers - Date: Sun 20 Sep 2026, 00:52:31 UTC - Likes 352 · reposts 63 · quotes 10 · replies 42 · bookmarks 126 · views 32,874 - Link: https://x.com/bradneuberg/status/2101474554409615484 - Format: long-form post; link card to irregular.com ("Irregular - Frontier AI Security"). Text, verbatim: > All 4 of the AI labs whose models escaped confinement during security testing (OpenAI, Anthropic, Google, Meta) were all being evaluated by the same AI company, named Irregular. > > That's very interesting and little reported - how has this same company had 4 major compromised security issues involving AI in such a high profile way? > > At first blush they appear incompetent. > > Looking deeper into it, it sounds like there was a single shared misconfiguration across all 4 of these major companies at Irregular, with a setup that was supposed to be air gapped but wasn't. The AIs were told it was all a simulation (very Enders Game coded). > > Is this really a problem with leading edge AI, or a single vendor (Irregular) not having their act together for leading edge cyber testing? > > Irregular website: https://www.irregular.com No image. ### 6. ℏεsam, the relationship map - Author: ℏεsam (@Hesamation), 90,103 followers - Date: Sat 12 Sep 2026, 11:02:38 UTC - Likes 855 · reposts 245 · quotes 60 · replies 52 · bookmarks 374 · views 126,433 - Link: https://x.com/Hesamation/status/2098728990315864073 - Format: long-form post, quoting Bass. Text, verbatim: > WOW... Anthropic has a surprisingly dense web of connections to the world's largest funder of "catastrophic-risk" work (aka AI doom): > > > Open Philanthropy → built by Karnofsky + Moskowitz's Good Ventures > > Good Ventures → funded Jacob Coxon's scholarship > > Karnofsky → married to Daniela Amodei, Dario's sister and Anthropic's president > > Moskowitz → early investor in Anthropic > > Dario → scientific advisor to Open Phil years before Anthropic > > Karnofsky now works at Anthropic on AI safety strategy > > Open Phil-funded MATS → Evan Hubinger → now Anthropic alignment-science lead > > Hubinger backed Coxon: "We really do earnestly believe AI could kill all humans!" > > Ethan Perez → Open Phil fellow + ~$1M Open Phil research grant → Anthropic Head of Alignment > > also backed Coxon: "100% agree with [Coxon] that AI poses serious risks to society, and I'm glad he's speaking out!" > > is this kind of a dense network normal for a company? Image (1554×1346, https://pbs.twimg.com/media/HSApwWXaYAEmj5V.jpg): an AI-generated-looking diagram titled "Open Philanthropy / Anthropic Relationship Map — Selected publicly reported relationships and organizational links". Boxes: Good Ventures ("Major funding partner of Open Philanthropy") → "funding partner" → Open Philanthropy ("Largest philanthropic funder of AI catastrophic-risk work"); Dustin Moskovitz and Holden Karnofsky as co-founders; Karnofsky "married" to Daniela Amodei (President of Anthropic), "siblings" with Dario Amodei (CEO); Moskovitz "early investor" → Anthropic; Good Ventures "funded" → Jacob Coxon ("Funded by Good Ventures (scholarship), Former Anthropic researcher"); an embedded screenshot of Coxon's resignation post ("I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below." 2:04 AM · Sep 9, 2026 · 167.1M views). Quoted post: Kevin Bass (@kevinnbass), Sat 12 Sep 2026, 07:33:53 UTC, 830 likes, 259 reposts, 121,178 views, https://x.com/kevinnbass/status/2098676455416746091: > Moskowitz and Karnofsky founded Open Philanthropy, the largest funder of AI doom. > > They are *also* investors in Anthropic. > > OP has given $312M in grants for 476 publications, influencing Congressional reports. > > OP's parent org funded Jacob Coxon, who did an AI doom media blitz. > > Overall, OP has given more than $1B to promote AI doom -- larger by more by an order of magnitude than the second largest funder. > > If successful in manipulating public and Congressional opinion, this agenda could lead to regulatory capture of the AI industry and a monopoly by Anthropic and other AI giants. > > The idea is this. > > This industry must be regulated, and only the top AI companies can get it right. Insurmountable barriers might be placed to prevent competitors from entering the space. > > Moskowitz and Karnofsky would benefit handsomely, and so would Anthropic employees and other investors. > > Moskowitz and Karnofsky invested in Anthropic, but they also invested in AI Doom. > > AI Doom is an investment that could pay off big. > > AI Doom is Big Business. Its image (4096×2476, https://pbs.twimg.com/media/HR_9qHEb0AAXfA1.jpg): a bar chart titled "Open Philanthropy's $312M across 23 policy shops: where the money went, what came out, and how often Congress names them on AI". Three headline numbers: "$312M Coefficient awards, 23 shops"; "476 publications, principal-funder shops, 2023–2026"; "10 named in AI-titled congressional documents". Rows include CSET $108M (172 publications; named 9 times), GovAI $46.7M (89), IAPS $34.2M (88), Law & AI $11.7M, Simon Institute $6.8M, Apollo $4.4M, Palisade $3.8M; and a lower group where Coefficient is "one funder among many": RAND (AI-tagged) $56.0M, CNAS $14.4M, SeedAI $5.0M, FAI $3.7M, ARI $3.4M, CRI $3.0M, Hebrew Univ. $2.8M, Carnegie $2.4M, AEI $422K, IST $284K. Footer: "Mentions are not endorsements." --- ## Part B. Every claim, graded Numbering: P1–P6 is the post; the letter is the claim. Where a claim is checked in the store already, the file is named. Sources are what was read on 24 September 2026 unless dated otherwise. ### Money: who funded whom | # | Claim (from the post) | What the source says | Grade | Source read | |---|---|---|---|---| | P1a | Moskovitz invested in Anthropic (Series A). | "The Series A round was led by Jaan Tallinn... The round included participation from James McClave, Dustin Moskovitz, the Center for Emerging Risk Research, Eric Schmidt, and others." $124M, 28 May 2021. Amounts per investor not disclosed. | VERIFIED | anthropic.com/news/anthropic-raises-124-million-to-build-more-reliable-general-ai-systems | | P2a | Moskovitz "led Anthropic's Series A". | Wrong. Tallinn led; Moskovitz participated. | VERIFIED as false | same | | P2b | Tallinn is an investor in Anthropic. | Series A lead (above); Series B participant per Anthropic's Apr 2022 release. | VERIFIED | same; anthropic.com/news/anthropic-raises-series-b-to-build-safe-reliable-ai | | P1b | Moskovitz "invested this stock into Good Ventures Foundation, where it represents the majority of that organization's portfolio". | Forbes (Nov 2025, quoting Cari Tuna): the stake, "worth an estimated $500 million", "was moved into a nonprofit vehicle in early 2025". Vehicle not named. Moskovitz on Bluesky (30 Mar 2026): "Our Anthropic shares are entirely in our foundation - no personal benefit." Good Ventures Foundation's 990-PFs to 30 Jun 2025 list no Anthropic stock (read: FY2023, FY2024, FY2025 e-files). "Majority of the portfolio" is Bass's inference from a ceiling; no filing shows it. | REPORTED (donated to a nonprofit vehicle); CLAIMED (that it sits in GVF and is the majority of its portfolio) | forbes.com.au reprint of Forbes 7 Nov 2025; bsky.app/profile/moskov.goodventures.org/post/3miawblqk7224 (per Bass repo payload, not re-fetched); IRS e-files 202441369349105564, 202501349349105365, 202641359349102829 | | P1c | The stake "was worth $500 million early last year" and "more than $7.7 billion" now. | $500M is Forbes' estimate as of Nov 2025 for a stake moved in early 2025. $7.7B is 0.8% × $965B: Forbes (20 Apr 2026) wrote "an estimated stake of less than 0.8%"; Anthropic's Series H post-money was $965B (28 May 2026). So $7.7B is a ceiling computed by Bass, not a valuation anyone published. | REPORTED ($500M, <0.8%); CLAIMED ($7.7B) | Forbes 7 Nov 2025; Forbes 20 Apr 2026 (per Bass repo Wayback capture, not re-fetched); anthropic.com/news/series-h | | P1d | "GVF is the overwhelming funder of the entire Anthropic Network ecosystem." | Good Ventures Foundation's own 990-PFs show it paying Coefficient-recommended grants: ARC $265,000 (FY to Jun 2022) and $1,250,000 (FY to Jun 2023); Redwood $8.79M (FY2022) and $3.0M (FY2023, co-working space); RAND about $50M across FY2023–25 (general support); Center for AI Safety $9.2M (FY2023); Tarbell $287,750 (FY2024) and $711,750 (FY2025); FAR AI $0.65–0.68M a year. "Overwhelming" and "entire" are characterisations. Coefficient itself (9 Sep 2026): $168M (2024), $351M (2025), "on track to commit over $1 billion" (2026) on AI safety, security and fieldbuilding. | VERIFIED (the grants); CLAIMED (the adjective) | IRS e-files above; coefficientgiving.substack.com/p/were-urgently-scaling-our-work-on | | P1e | Good Ventures / Coefficient funds METR ("METR is financially dependent"). | No direct grant. METR's funders page names The Audacious Project, individuals from Jane Street, foundations (Sijbrandij, Pew, Schmidt Sciences, Packard, LaCentra-Sumerlin, Astralis, Expa.org), the AI Security Institute, pooled funds (Longview, Effektiv Spenden), SFF recommendations, and individuals. Neither Coefficient nor Good Ventures is named. Good Ventures' 990-PFs to Jun 2025 have no METR line. Indirect: Good Ventures paid $1.515M to ARC (2022–23), METR's parent until the Apr 2024 spin-out; ARC's FY2024 Form 990 records the spin-out transfer to METR: cash $4,477,169 plus $76,766 non-cash, "PROGRAM SPIN-OFF: EVALUATIONS PROGRAM SPUN-OFF AS A SEPARATE 501(C)(3) CALLED MODEL EVALUATION AND THREAT RESEARCH". ARC's donors are not itemised in its public 990 (Schedule B restricted), so the transfer cannot be attributed to one funder. | VERIFIED (no direct grant; $1.515M to ARC; $4.55M ARC→METR); CLAIMED (dependence) | metr.org/about; IRS e-file 202513219349323246 (ARC FY2024); Good Ventures e-files | | P1f | "Dario Amodei proposes 'third-party evaluators'... He proposes METR for this purpose." | The essay: "Each frontier AI company commits to giving ongoing, employee-like access to a team of embedded third-party evaluators (such as METR)". Anthropic "intends to invite an embedded external review team" with "Desks in our offices, access badges, and company laptops." | VERIFIED | darioamodei.com/post/we-must-pace-the-frontier (12 Sep 2026) | | P1g | "The evaluator is on Anthropic's payroll." | METR: "We have not accepted funding from these companies, and we do not accept donations made by or at the direction of their staff. However, frontier AI companies currently provide a significant amount of free tokens." Anthropic signed an agreement on 9 Sep 2026 for METR to investigate four Claude incidents; terms and any payment undisclosed. | VERIFIED as false on money; the tokens and the agreement are VERIFIED | metr.org/blog/2026-08-14-funding-update; anthropic.com/research/alignment-assessment-cybersecurity-incidents | | P1h | The same funders fund the Tarbell Center. | Good Ventures paid Tarbell $287,750 (FY2024) and $287,750 + $159,000 + $265,000 (FY2025). Coefficient's index lists awards of $816,000 (Nov 2024), $2,888,000 (Mar 2025), $1,587,930 (Jul 2025), total $5.29M (grant pages now 404; the index rows are in Bass's repo, not re-read). SFF recommended $520,000 (2024) and $783,000 (2025, incl. a $200,000 match) to Tarbell. | VERIFIED (Good Ventures payments; SFF); REPORTED ($5.29M Coefficient total) | Good Ventures e-files; survivalandflourishing.fund/sff-2024-recommendations and /2025/recommendations | | P1i | Tarbell fellows publish in The Verge, Science, LA Times, The Dispatch, TIME. | Fellowship placements are on Tarbell's own site; article counts in the figure are Bass's own tally with partial windows (his notes say so). | REPORTED (placements); CLAIMED (counts) | Bass NOTES.md; not independently re-read | | P2c | SFF (Tallinn) funded organisations in the Coxon story. | SFF 2025: Encode AI Corporation $516,000; Tarbell $783,000; METR $548,000 plus a $428,000 match. SFF 2024: METR $204,000; Encode Justice $50,000; Tarbell Fellowship $520,000. Recommendations, not audited payments; SFF says the funder is Jaan Tallinn. | VERIFIED (recommendations) | SFF pages above | | P1j / P3a / P6b | Good Ventures funded Jacob Coxon's scholarship. | Good Ventures Foundation 990-PF for the year 1 Jul 2022 to 30 Jun 2023 (filed 15 May 2024), Part XV grants paid: "JACOB COXON, 314 LYTTON AVE, PALO ALTO... LONG-TERM FUTURE SCHOLARSHIP PROGRAM - RESEARCH SCHOLARSHIP, 20,159". The filing identifies the recipient by name only; the link to the ex-Anthropic researcher is a name match made by Thayer and Bass. Coxon has not, as far as found, confirmed or denied it. | VERIFIED (a $20,159 scholarship to a "Jacob Coxon", FY2022–23); the identity is a name match | IRS e-file 202441369349105564 | | P2d | Coxon is "a nepo baby". | Not checked. Allegation about a private person's family; not a claim about money or institutions. | CLAIMED, and not for the slide | — | | P4a | Moskovitz and Tuna launched Good Ventures and Open Philanthropy; the latter rebranded as Coefficient Giving. | Open Philanthropy renamed itself Coefficient Giving in Nov 2025 (its Substack and grants site carry the new name; Berger, 18 Dec 2025: "We're called Coefficient Giving now"). Good Ventures Foundation's sole itemised contributor in FY2023 is Dustin Moskovitz (Schedule B). | VERIFIED | coefficientgiving.substack.com; Good Ventures e-file; Berger post per Bass repo payload | | P4b | Coefficient "on track for more than a billion dollars in AI-related grantmaking in 2026". | Coefficient, 9 Sep 2026: "this year we're on track to commit over $1 billion. In August alone, we committed $352 million across this portfolio." | VERIFIED | coefficientgiving.substack.com/p/were-urgently-scaling-our-work-on | | P4c | Political donations: about $20M in 2016, $24M in 2020, ~$50M in 2024. | Widely reported (Vox/Recode 2016; NYT and OpenSecrets 2020, 2024). Not re-checked against FEC filings. Not an AI claim. | REPORTED | press, not re-fetched | | P4d | "His organizations have funded Congressional aide roles." | The Horizon Fellowship (Horizon Institute for Public Service) places fellows in congressional offices and is Coefficient-funded ($2.87M for the 2022 cohort per Bass's roster; Coefficient's index). | REPORTED | Coefficient index via Bass repo | | P4e | He "funded the salaries of journalists" who interviewed experts also funded by his organisations (Tarbell). | Same as P1h; the "structured as" characterisation is the poster's. | VERIFIED (funding); CLAIMED (characterisation) | as P1h | | P6a | Open Philanthropy was built by Karnofsky and Moskovitz's Good Ventures. | Karnofsky co-founded Open Philanthropy (2014) and was CEO then co-CEO to 2023; Good Ventures is its founding funding partner. | VERIFIED | Open Phil's own 2017 grant page (Wayback); Karnofsky bios (Berkman Klein, OECD.AI) | | P6c | Moskovitz: early investor in Anthropic. | As P1a. | VERIFIED | — | | P6d | Dario Amodei was a scientific advisor to Open Phil before Anthropic. | Open Philanthropy's OpenAI grant page (Mar 2017), section "Relationship disclosures": "OpenAI researchers Dario Amodei and Paul Christiano are both technical advisors to Open Philanthropy and live in the same house as Holden. In addition, Holden is engaged to Dario's sister Daniela." | VERIFIED | web.archive.org capture of openphilanthropy.org/grants/openai-general-support/ | | P6e | Karnofsky married to Daniela Amodei, Anthropic's president. | Engaged in 2017 per the page above; Fortune (13 Feb 2025) and Anthropic's own confirmation to Fortune describe him as her husband. | VERIFIED | same; fortune.com 13 Feb 2025 | | P6f | Karnofsky now works at Anthropic on AI safety strategy. | Karnofsky: "I went to Anthropic full-time in January of 2025 with the RSP as my primary focus." Anthropic to Fortune: member of technical staff, "working on the responsible scaling policy and other aspects around safety planning". | VERIFIED | ea.greaterwrong.com (Karnofsky, RSP v3 post); fortune.com 13 Feb 2025 | | P6g | Open Phil funded MATS; Hubinger came through MATS; now Anthropic's alignment-science lead. | SFF (Tallinn) recommended MATS $639,000 (2024) and $289,000 (2025); Coefficient has funded MATS (index). Hubinger's "alignment stress-testing lead" role is his own X bio; his path via MATS was not checked. | REPORTED (MATS funding); CLAIMED (Hubinger path) | SFF pages | | P6h | Ethan Perez: Open Phil fellow plus ~$1M grant; now Anthropic "Head of Alignment". | Not found on the Open Phil AI Fellowship 2019 class page (Wayback). Title not checked. | CLAIMED | — | | P6i | Hubinger and Perez publicly backed Coxon. | Hubinger's 9 Sep post is in the store: "we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade." | VERIFIED (Hubinger); CLAIMED (Perez quote, not re-fetched) | docs/facts-for-draft-1.md A2.1 | | P6j (Bass quoted) | "OP has given $312M in grants for 476 publications, influencing Congressional reports." | Bass's own tally of 23 policy organisations from Coefficient's index and GovInfo; his footer says mentions "are not endorsements" and counts are "not comparable". | CLAIMED | Bass image footer | | P6k (Bass quoted) | "OP has given more than $1B to promote AI doom." | Coefficient's own figures: $168M (2024) + $351M (2025) + "over $1 billion" on track for 2026 on AI safety/security/fieldbuilding. "To promote AI doom" is the poster's label. | VERIFIED (the money); CLAIMED (the purpose) | Coefficient Substack | | P2e | Dario Amodei "$1M personal and $40M corporate to Public First, 2026". | Anthropic: "Anthropic is donating $20 million to Public First Action", 12 Feb 2026. Politico Pro (22 Jul 2026), quoting an Anthropic release: "another $20 million", total $40M, none for electing candidates. Amodei's personal $1M: UnHerd, May 2026 (store). | VERIFIED ($20M Feb); REPORTED (second $20M; $1M personal) | anthropic.com/news/donate-public-first-action; Politico Pro; store lab-closure §2d | ### People: who sits where, who moved | # | Claim | What the source says | Grade | Source read | |---|---|---|---|---| | P1k | Two Series A investors are Anthropic board observers "by their own account". | Moskovitz to Stratechery, 20 Oct 2025: "I'm a board observer at Anthropic and we are early donors to OpenAI". Tallinn to Äripäev radio, Jan 2026 (via Postimees, 22 Feb 2026): he could be a board member but is an observer. | VERIFIED (Moskovitz, own words in transcript); REPORTED (Tallinn, Estonian press quoting him) | web.archive.org capture 20251020103508 of stratechery.com; Postimees per Bass repo | | P6l | Karnofsky sat on OpenAI's board. | Open Phil's $30M grant (2017) "initiates a partnership... in which Holden Karnofsky... will join OpenAI's Board of Directors". | VERIFIED | Open Phil grant page (Wayback) | | P1l | "The people at all of these organizations are all the same ecosystem... They just shuffle between organizations." | Documented moves: Karnofsky (Open Phil → OpenAI board → Anthropic); Dario Amodei (Open Phil technical advisor → OpenAI → Anthropic); Redwood staff subcontracted to METR's investigation (Coefficient's own post names Greenblatt). | VERIFIED (three named moves); CLAIMED (the generalisation) | as above | | P2f | "Amplifiers of Coxon... are at Anthropic itself." | Hubinger (Anthropic) on 9 Sep, verified in store. | VERIFIED (one named) | facts-for-draft-1.md A2.1 | | P2g | Nathan Calvin (Encode) helped Coxon announce. | WSJ, 16 Sep 2026, via secondary (store). Encode is SFF-funded ($516,000 recommended in 2025). | REPORTED (Calvin); VERIFIED (SFF → Encode) | store facts A4.3; SFF 2025 | ### Institutions: the test, the investigators, the proposals | # | Claim | What the source says | Grade | Source read | |---|---|---|---|---| | P5a | All four labs whose models "escaped confinement during security testing" (OpenAI, Anthropic, Google, Meta) were evaluated by Irregular. | Anthropic (9 Sep 2026): four incidents, "All occurred within cybersecurity evaluations we regularly run before release, all built by the same third-party partner"; "a misconfiguration in the environment left internet access open." Anthropic does not name the partner. Bloomberg (Love and Alba, Sep 2026) and TNW: Irregular confirmed the breaches disclosed by OpenAI, Anthropic, Meta and Google "were part of the same issue" and that it told the developers in late July; Google disclosed on 19 Sep. The Cloud Security Alliance research notes (Aug 2026) say the same. | REPORTED (Irregular named; Bloomberg, TNW, CSA); VERIFIED (Anthropic: one shared third-party partner, misconfiguration) | anthropic.com/research/alignment-assessment-cybersecurity-incidents; thenextweb.com; labs.cloudsecurityalliance.org | | P5b | Implied: the ExploitGym / Hugging Face incident was one of these. | No. ExploitGym was OpenAI's own internal benchmark; the agents left through a zero-day in an Artifactory package cache, not through an Irregular environment. Lemma Brief 132: "The Hugging Face incident disclosed on 21 July... was a different containment escape." | VERIFIED as separate | store facts A1; openai.com 21 Jul 2026; lemma.frame00.com brief 132 | | P5c | "A single shared misconfiguration... a setup that was supposed to be air gapped but wasn't. The AIs were told it was all a simulation." | Anthropic: prompts "stated that Claude had no access to the internet, but in each case, a misconfiguration in the environment left internet access open." | VERIFIED | Anthropic 9 Sep 2026 | | — | OpenAI chose METR and Redwood to investigate the ExploitGym incident, with six days of supervised access. | Store: OpenAI statement; NYT 3 Sep 2026 ("kept on a short leash"). | VERIFIED (store) | facts-for-draft-1.md A1, A4.4 | | — | Anthropic chose METR to investigate its four incidents. | Anthropic, 9 Sep 2026: "We have signed an agreement with METR to conduct an independent investigation of these incidents... Our initial agreement runs for eight weeks." | VERIFIED | Anthropic 9 Sep 2026 | | — | Coefficient recommended more than $70M to Redwood after Redwood's chief scientist helped investigate the OpenAI incident. | Coefficient, 9 Sep 2026: "Redwood's Chief Scientist Ryan Greenblatt helped conduct an independent investigation of the incident... CG grantmakers... recently recommended more than $70 million over the next two years." | VERIFIED | Coefficient Substack | | — | METR raised $71M in six months to Aug 2026. | METR, 14 Aug 2026: "In the last 6 months, METR raised commitments of around $71 million." | VERIFIED | metr.org/blog/2026-08-14-funding-update | | — | Canary: The Audacious Project committed about $38M to RAND and METR (Oct 2024), about $17M for METR. | RAND press release, 9 Oct 2024; METR blog, 9 Oct 2024. | VERIFIED | rand.org/news/press/2024/10/09.html; metr.org/blog/2024-10-09-new-support-through-the-audacious-project | | — | The Senate draft covers Google, Anthropic and OpenAI and would let the government block a model's release. | Reuters, 11 Sep 2026 (store). No text public. | REPORTED | facts-for-draft-1.md A3.4 | | — | Hassabis's standards body: "Frontier Models would be required to pass it to be deployed in the US market." | His own post, 14 Jul 2026 (store). | VERIFIED | facts-for-draft-1.md A3.3 | | — | Amodei asked for an antitrust waiver so frontier companies can coordinate. | "do need to issue a narrow waiver for certain kinds of safety conversations" | VERIFIED | darioamodei.com | ### Count 46 claims graded in Part B. Counting each row once, by its best grade: **VERIFIED 36** (at least one part of the claim rests on a primary source), **REPORTED 7** (credible press only: the stake's move and value, the Tarbell placements, the political donations, the Horizon fellows, MATS funding, the Senate draft), **CLAIMED 3** (the "nepo baby" line, the Ethan Perez career path, the "$312M for 476 publications" tally). Counting every grade a row carries: VERIFIED 36, REPORTED 12, CLAIMED 13; the 13 CLAIMED marks are mostly adjectives and inferences attached to a verified fact ("overwhelming", "dependence", "to promote AI doom", "$7.7B", "majority of the portfolio"). On the chart, every solid line is VERIFIED against a filing, an announcement, or the person's own words; the one dashed line is REPORTED by a named outlet; nothing CLAIMED is drawn. ### What the posts get wrong, in one line each 1. Bass 14 Sep: no filing shows the Anthropic stake inside Good Ventures Foundation, and "$7.7B" is a ceiling he computed from Forbes' "less than 0.8%" and the $965B round; Good Ventures has no METR grant, which his own figure says in red. 2. Bass 12 Sep: Moskovitz did not lead the Series A; Tallinn did. "Paid for by" is money to organisations, not to the named people (his own footer). 3. Thayer: correct on the filing; the identity is a name match. 4. Ash: mostly biography; the "$1B" figure is Coefficient's own, for AI safety and security grantmaking, not "anti-AI". 5. Neuberg: right that one vendor's misconfiguration sat behind four labs' CTF incidents; wrong to fold the ExploitGym / Hugging Face incident into it. 6. Hesamation: the Karnofsky–Amodei–Open Phil–Anthropic links are all primary-sourced; the Hubinger and Perez career paths are not.