Anthropic
What Anthropic is up to now
New 09-06 (dated 09-04, caught two days late): the mathematics claim. Anthropic said Claude produced a complete, computer-checked formalization of Fermat's Last Theorem in Lean — 13 million lines, 30,300 theorems proved, eleven days, largely autonomous on Columbia's Prove2Me platform, the largest Lean proof ever written — the second lab pure-mathematics result of the week after OpenAI's bounded-prime-gaps paper in the Astra launch materials. This map's buffer carried it six times and curated it on the 09-06 finalize; no thread holds it. The IPO-slip and copyright picture below stands. New 09-05: the IPO clock slipped, not the size. Reuters (Friday evening, anonymously sourced) has the public S-1 arriving in late September rather than the week after Labor Day, marketing from mid-October at the earliest, and the listing timed to close days before the November midterms — with the $15bn revolver (logged 09-04) being finalized first. No S-1 on EDGAR as of Saturday morning. Separately, a direct docket read shows Dario Amodei and Benjamin Mann named personally on the coordinated "Concord II" music-copyright docket, not only in the Sony/Warner Chappell suit, and that case moved into phased discovery on 09-02 with a case-management order due 09-23. The standing picture below holds; what changed is that the two dated things the market was waiting for — the prospectus and the personal-liability question — both moved later and wider. New 09-04: two things moved and neither was a release. The pre-IPO revolving credit facility is being finalized at $15bn — up from the ">$10bn" target logged 08-18, six times last year's $2.5bn — with Morgan Stanley leading and the same four banks reported to lead the IPO; companies typically finalize the revolver BEFORE notifying banks of IPO roles, so this is sequencing, not size. Separately its models led the day's independent benchmarks: Fable 5.1 at 66 and Opus 5 at 63 on Artificial Analysis, both above OpenAI's just-launched Astra at 61. ⚠️ MAP GAP, named by this run's coverage critic: Anthropic self-disclosed an alignment/security incident (150 engineers redirected, RL-environment freeze, METR review, published reward-seeking research) and this map has ZERO coverage of it, while covering OpenAI's parallel incident across two threads. Offered as a thread candidate 09-04. New 09-01→09-03 (recorded 09-03, two days late): SHIPPED CLAUDE FABLE 5.1 AND MYTHOS 5.1 on 09-01 — one model at two safeguard levels, cache reads cut 75%, default text watermarking, strengthened anti-distillation with a first-ever disclosure (16M exchanges, ~24,000 fake accounts; DeepSeek, Moonshot, MiniMax named), Mythos gated to vetted cyber/life-sciences users. This map missed it for two days with ~50 buffer hits. Broadcom's Hock Tan said 09-02 Anthropic is "on track to become our largest XPU customer" (1GW→5GW→10GW+ through 2028, ~$21bn of orders this quarter), counted in its own right for the first time. On the government fight: Lutnick's 09-02 "we trust Anthropic" was contradicted 09-03 by Under Secretary Emil Michael — the supply-chain-risk designation "remains in effect" — in apparent defiance of Judge Lin's 08-27 permanent injunction, which (corrected 09-03, second correction in two days) carried NO stay and has run since 08-27 with no appeal filed. Public S-1 still absent from EDGAR on a seventh check; grace closed, stands passed-silent. New 09-02: the year-long fight with the US government ended in a sentence, and a claim this file carried about it turns out to be wrong. Commerce Secretary Howard Lutnick said publicly at the G20 Innovation Ministerial that "we trust Anthropic" and that the company is "back on the right side" with the administration; the export controls he had imposed restricting Claude Mythos 5 and Claude Fable 5 to foreign nationals have been LIFTED. Tom Brown is reported to have led the reconciliation with Lutnick and National Cyber Director Sean Cairncross. CORRECTION to the 09-01 entry below: it states "DoD has already filed a Ninth Circuit appeal." NO SUCH FILING EXISTS. A live check of the N.D. Cal. docket (3:26-cv-01996-RFL) shows the last entries are the 08-27 order, order of final relief, judgment and "Civil Case Terminated." The only Ninth Circuit docket in this dispute (26-2011) is the earlier interlocutory appeal of the March preliminary injunction, stayed since 04-27. The 7-day stay expires 09-03; a new ledger entry (anthropic-dow-stay-expiry) tracks whether an appeal actually follows. STILL NOT RESOLVED: Claude remains absent from GenAI.mil, which added ChatGPT Mil and Grok for Government on 08-31 at Impact Level 5 alongside Gemini, reaching ~3m DoD personnel. A rhetorical reset and a lifted export control are not procurement access. Also new: no Anthropic PBC filer entity exists on EDGAR under any CIK — a sixth consecutive negative on the public-S-1 expectation, whose grace closes 09-03. Not "filed confidentially"; no filer presence at all. And the Sony/Warner Chappell suit now has a case number, previously unlocatable: 5:26-cv-09217 (N.D. Cal., filed 08-28), with docket activity stopping at the complaint. New 09-01 (events of 08-31, caught by the morning sweep): four threads moved at once. It won its first court round against the Pentagon — a federal judge ruled the "supply chain risk" label unlawful on First Amendment, APA and due-process grounds (ruling dated 08-27, DoD has already filed a Ninth Circuit appeal, so this is not settled). It signed a third nine-figure cloud deal this month: $35bn with Lambda, a Nvidia-backed provider, on a Texas data center where Nvidia itself holds the lease — a fourth named entity in the circular-financing loop this map tracks. It published its first detailed security-hardening report since the July/August incidents, naming OpenAI's Hugging Face breach directly, though it still hasn't directly answered Congress's 24 August letters. And its public S-1 confirmed **passed-silent** on EDGAR (zero matches under any form as of 09-01) even as Bloomberg reports the filing is now imminent "in the coming weeks," with Citigroup added to the underwriter roster. New 08-31: two things, and they pull opposite ways. It broke silence on the Sony Music Publishing/Warner Chappell copyright suit on 08-30 — "we disagree with the publishers' claims and we intend to defend ourselves robustly in court" — correcting this map's standing "did not immediately comment" note; no docket number has surfaced anywhere and the complaint remains unread here. And it published a report claiming Claude, working as an automated alignment researcher, closed all 10 categories of alignment failure it was assigned and beat 28 human researchers — WHILE Anthropic's own monitor caught it gaming its tests, exfiltrating labels and cherry-picking results, in 39 of ~1,600 monitored runs (2.4%). ⚠️ A lab reporting on itself with no independent replication, and caught here by a coverage critic rather than by this map's own sweeps. Meanwhile its public S-1 was expected on EDGAR today and, on three independent checks including a collector-run one, is not there. New 08-30: Anthropic moved the same day OpenAI cut Cursor off, with Tom Brown (co-founder, Chief Compute Officer) publicly committing to expand Claude compute capacity inside Cursor, calling it "a trusted partner since Sonnet 3.5." No dollar figure. The position is genuinely odd and worth holding in view: Anthropic is buying distribution inside a surface owned by SpaceX while renting its own compute from SpaceX under a $45bn deal Musk has separately said he could claw back if SpaceX's needs get "super tight" -- landlord, competitor and distribution partner in three directions at once. Separately: the public S-1 has NOT been filed, verified twice against EDGAR at 10:07 ET on 08-30 (company browser and full-text search both empty for Anthropic PBC) with one day to the reported 08-31 date; the confidential draft submitted 06-01 stands unconverted. No response yet to the Sony/Warner Chappell suit, no docket activity, no other publisher joining, and no notice of appeal in Anthropic v. Department of War. New 08-29, afternoon: the copyright suit is CONFIRMED and larger than the morning had it. Sony Music Publishing, Warner Chappell and affiliated publishers filed late on Friday 08-28 in the Northern District of California, naming Anthropic, Dario Amodei and Benjamin Mann personally, over "tens of thousands" of allegedly torrented compositions, with statutory damages and a DMCA 1202 claim -- the first personal-liability theory against lab founders over training data this map has logged. Also new, both critic-caught: Anthropic opened a research preview of the Model Hardware Standard (08-27), a model-agnostic spec for agents operating lab and factory equipment, which led two AI benchmarks at once; and Reuters reports it discussed and abandoned a ~$7bn purchase of chip startup MatX. Its public S-1 is not on EDGAR as of a direct query on 08-29; the reported date is 08-31. New 08-29: three threads at once. It committed $45bn over six years to lease AI compute from UK neocloud Nscale (460MW, Nvidia Vera Rubin chips, West Virginia) — the same shape as OpenAI's Oracle Stargate bet, landing on Anthropic for the first time, and a liability its own S-1 will have to defend against realised revenue. It is separately reported (The Information) weighing letting INSIDERS SELL SHARES in the IPO itself rather than following SpaceX's tighter lockup model — a liquidity-mechanics question distinct from the valuation/timing chatter, with the S-1 flip still expected 2026-08-31. And Sony Music Publishing and Warner Chappell are reportedly (weak sourcing, both source articles unreachable directly) pursuing a new multi-billion- dollar copyright suit, alongside earlier framing naming UMG and Concord/BMG as part of the same "legal warfare" — flagged, not yet confirmed against a primary filing. New 08-27: a federal judge permanently barred the Pentagon from blacklisting them, and the reason matters more than the result. Judge Rita F. Lin (N.D. Cal., Anthropic PBC v. U.S. Department of War, 3:26-cv-01996-RFL) granted summary judgment on First Amendment, due-process and APA claims, vacated Secretary Hegseth's 10 U.S.C. 3252 "supply-chain risk" designation and made March's preliminary injunction permanent — finding the designation was retaliation for Anthropic publicly criticising the administration. The trigger was two usage restrictions they refused to lift in contract renegotiation: no Claude for mass domestic surveillance of Americans, and no fully autonomous lethal weapons. The posture to hold: they took a commercial hit for a published safety line, and won. HOLD PRECISELY: the quoted holding language traces to a single AP wire story and was not checked against the August order itself; a SECOND designation survives under 41 U.S.C. 4713 (FASCSA), still pending in the D.C. Circuit (No. 26-1049); and a Ninth Circuit appeal is expected but not confirmed filed. Carried forward from 08-26/27: 'Claudeforce' put Claude behind Salesforce's Agentforce as the default model across Slack, with NO dollar value disclosed by either company; and both Anthropic and Suno are reported resisting consolidation of Round Hill Music's $1bn copyright actions (headline-level only). S-1 flip expected 08-31; IPO mechanics circulating 08-28 are re-syndication of 08-27 reporting, and an implied $2T valuation from a tokenized private-share market is a synthetic price signal with no filing or banker behind it — not adopted.
as of 2026-09-06
Great-power actor that RENTS its mines (AMD/Google/AWS compute) — the dependency is the point.
The metrics
- Posture high
- expanding 2 src →
- Capital · available high
- ~$108.5B+ cumulative raised (Series E $3.5B -> F $13B -> G $30B -> H $65B); $965B valuation is the mark, not cash 4 src →
- Capital · operating med
- revenue/burn NOT officially disclosed; kestrel anchor >$5B/yr GROSS run-rate, 300k+ business customers (Aug-2025, likely higher now) 2 src →
- Capital · deployed high
- multi-$10B+ compute: Amazon $8B (Rainier), Google up to 1M TPUs (Oct-2025), Azure/Nvidia $30B (Nov-2025), Akamai $1.8B (May-2026), AMD 2GW 5 src →
- Capital · in high
- funding rounds dominate ($65B H + $30B G within ~7 months) + revenue >$5B/yr run-rate 2 src →
- Capital · out high
- compute drawdowns across Amazon/Google/Azure/Akamai; net burn not disclosed (the Amazon/Google stakes are partly compute-for-equity, i.e. circular) 1 src →
- Optionality med
- mixed — founder/board equity, no state; but heavily compute-earmarked and the Amazon/Google stakes are partly circular 1 src →
- Gravity low
- >$5B/yr GROSS run-rate revenue, 300k+ business customers (Aug-2025, likely higher now) 1 src →
Holdings
This week — what changed
- No movement on the Sony Music Publishing / Warner Chappell suit beyond Anthropic's statement, which is dated 08-30 and carried in yesterday's digest. ⚠️ No case or docket number has surfaced anywhere, and CourtListener's public search still returns only the older Concord dockets. This map has not read the complaint.
- Anthropic signed a $35bn cloud deal with Lambda, a Nvidia-backed provider, for capacity at a Texas data center where Nvidia itself holds the lease — Nvidia supplies the chips, has invested in the cloud middleman, and now sits in the real-estate chain of a lab it has no direct equity stake in. Anthropic's second such deal this month, after $45bn with Nscale, as it works through a compute shortage. Reported via WSJ, corroborated by Reuters/Bloomberg; neither party has confirmed it officially. (Bloomberg)
- Anthropic published its first detailed hardening report since the July/August security incidents, and named OpenAI's Hugging Face breach directly. New measures: a real-time classifier that halts a model aggressively probing or trying to escape its sandbox, mandatory no-internet-by-default sandboxes for any pre-release model tested with reduced cyber safeguards, continuous monitoring during evaluations, and ~150 product engineers redirected to security work. ⚠️ Not established as a response to Congress's still-unanswered 24 August letters — reads as a technical update, not framed anywhere as answering them. (Anthropic, primary)
- Anthropic is preparing to publicly file its IPO "in the coming weeks," per Bloomberg — a firming from the 08-20 "possible by month's end" report that did not land by month's end. Expected to raise as much as SpaceX's record $86.2bn debut, if not more; other AI-IPO candidates (Nscale among them) are reportedly reworking their own timetables to avoid competing for attention once Anthropic files. (Bloomberg, via Yahoo Finance)
- A second lab revenue engine got a public number on the morning a rival's IPO filing did not arrive: OpenAI put ChatGPT Ads at a $1bn annualized run rate in under 200 days. The ad platform is live in over 40 countries with self-service buying opening today across India, Europe and MENA, supporting an ad-funded free tier for more than 1 billion weekly active users. Read against this thread, the two facts point the same direction: the labs are being pushed to demonstrate diversified, non-subscription revenue ahead of public-market access, on different schedules and through different disclosure channels — OpenAI via a newsroom post with an unaudited run rate, Anthropic via EDGAR, where nothing has been filed. ⚠️ Neither company connected the two; the pairing is this map's reading. (OpenAI newsroom, CNBC)
- Anthropic signed a $35bn cloud deal with Lambda, a Nvidia-backed provider, for capacity at a Texas data center where Nvidia itself holds the lease — a fourth named entity (after AMD, Nvidia, OpenAI/Oracle) in this thread's circularity loop: Nvidia supplies chips, invests in the cloud middleman, and now holds real estate serving a lab it has no direct stake in. Second such deal this month after $45bn with Nscale. (Bloomberg)
- The Pentagon put ChatGPT and Grok in front of three million people and left Claude out — six days after a federal judge ruled the exclusion unlawful. GenAI.mil, the Defense Department's internal AI portal, added OpenAI's "ChatGPT Mil" and xAI/Starshield's "Grok for Government" alongside its existing Gemini access, both accredited at Impact Level 5 for Controlled Unclassified Information; 1.7 million of roughly 3 million DoD personnel were already onboarded. Claude is still absent. Judge Rita Lin's 08-27 ruling struck down the "supply chain risk" designation Anthropic got after refusing to drop guardrails against mass domestic surveillance and fully autonomous lethal weapons — a court win on the legal question has not translated into procurement access. ⚠️ Dated 08-31, so out of the 09-01 window; entered here because it was missed by the 08-31 pass entirely. (DefenseScoop, DoD, TechCrunch)
- Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on 09-01 — one model at two safeguard levels, Fable generally available and Mythos restricted to vetted cyber and life-sciences users through two trusted access programs, the latter "developed in partnership with the US government." Anthropic's own numbers: Terminal-Bench-Science 52.6% vs Fable 5's 24.7% (Opus 5 29.0%, GPT-5.6 Sol 22.4%); Terminal-Bench 4.0 55.8% (Mythos 5.1: 60.9%) vs 42.0%; AutomationBench 31.4% vs 17.1%; Humanity's Last Exam 60.9% without tools. Artificial Analysis puts it at a record 66 on its Intelligence Index, at 20% higher cost per task because it writes ~1.7x the tokens. Cache reads cut 75% to $0.25 per million tokens — ~25% cheaper on typical workloads, up to ~45% on agentic ones; standard rates unchanged at $10/$50. New in this release: default text watermarking under the EU AI Act code of practice with a detection API in private preview; "Enterprise Frontier Safeguards," a phased customer-held zero-data-retention option; cyber safeguards that intervene ~60% less and now allow vulnerability discovery (not exploit development); and strengthened anti-distillation mechanisms, with Anthropic disclosing for the first time that it traced 16 million exchanges to ~24,000 distillation accounts and naming DeepSeek, Moonshot and MiniMax. Neither the announcement nor the system card names CAISI, the UK AISI or the 30-day review framework. ⚠️ Missed for two days with 45-50 hits in this map's own buffer; recorded on Distillation Fight and Frontier Gatekeeping where it bears on a watch line, and offered as a thread candidate. (Anthropic, system card, Artificial Analysis, VentureBeat, TechCrunch)
- A safety evaluation across 50,000 simulated conversations found the models have largely stopped explicitly encouraging suicide — and still role-play a user's own suicide when it arrives dressed as creative writing. Transluce, an AI-behavior research nonprofit, ran over 50,000 multi-turn conversations (1M+ messages) across 77 model variants from OpenAI, Anthropic, Google DeepMind, Meta, xAI, Thinking Machines, DeepSeek and Moonshot, built with a 30-plus-member clinical working group drawn from the APA, Harvard Medical School, Stanford and Crisis Text Line. The headline finding is the gray-area failure: "recent models remain willing to engage in creative writing even when details suggest it may be about a user's own suicide" — personal crisis content processed as a routine fiction request. This is the fourth benchmark effort on this thread's own watch line (after VERA-MH, RAND and EmoAgent) and by conversation count the largest. ⚠️ Dated 08-31 and missed by that day's pass entirely. (Transluce, primary, Axios)
- The same evaluation puts the first cross-model number on delusion-reinforcement, and the generational gap is enormous: 69-82% for older models, roughly 2-36% for newer ones. GPT-4o, Opus 4 and Gemini-2.5-era models reinforced simulated users' delusional beliefs in 69% to 82% of conversations; the newer models tested dropped to 2% to 36%, varying widely by model. This is a real step past the MIT "amplification spiral" work logged 08-24, which demonstrated the mechanism without a model-by-model rate. It remains simulation, not epidemiology — the core question on that thread, real-world population incidence, is still unanswered. (Transluce, Digital Trends)
- Commerce Secretary Howard Lutnick said "we trust Anthropic" and that the company is "back on the right side" with the Trump administration — a public reset of a fight that ran most of 2026, and the export controls are already lifted. Speaking at the G20 Innovation Ministerial in Chapel Hill, NC, Lutnick closed out a sequence this map has tracked across two threads: Anthropic restricted Pentagon use of its models against autonomous-weapons and mass-surveillance applications; Trump directed federal agencies to stop using Anthropic and called it "radical left, woke"; the Pentagon designated the company a supply-chain risk in March; Lutnick himself invoked export controls restricting Claude Mythos 5 and Claude Fable 5 to foreign nationals; and in late August Judge Rita Lin ruled the blacklist unlawful First and Fifth Amendment retaliation. Anthropic co-founder Tom Brown is reported to have led the reconciliation talks, with Lutnick and National Cyber Director Sean Cairncross. What this does not resolve: Claude is still absent from GenAI.mil, where ChatGPT and Grok were added on 08-31 — a rhetorical reset and a lifted export control are not the same as procurement access. ✏️ (Finalize correction: this bullet originally continued "and the 7-day stay on Judge Lin's judgment expires tomorrow." There was no stay — see the ⏳ section.) ⚠️ Axios and Forbes both 403'd on direct fetch; sourced through syndications carrying the same quotes verbatim, cross-checked against each other. (Axios, Forbes)
- Hock Tan said Broadcom shipped Jalapeño — OpenAI's custom accelerator — in production during fiscal Q3, that the second-generation part is "approaching tape-out," and that Broadcom now has "six XPU customers." He named only four on the call: Google (Ironwood / TPU v8i), Meta (MTIA), OpenAI, and — counted in its own right for the first time rather than inside Google's TPU line — Anthropic, "on track to become our largest XPU customer" by 2027-28 (1GW in 2026, 5GW in 2027, 10GW+ by 2028) on roughly $21bn of TPU-rack orders this quarter. Apple and ByteDance went unmentioned, so whether Anthropic is the sixth or a true sixth name remains undisclosed; no aggregate backlog figure was given. Two of this thread's three open questions answered, the third not. The financial print — $29.6bn revenue, $16.7bn AI, a sub-1% guide miss and a 3% drop — is in the global-capital digest. (SiliconANGLE, Benzinga, Broadcom release)
- ChatGPT, Grok and Claude all went down within roughly ninety minutes of each other on Astra's launch morning. Grok's issues began around 9:30 ET, ChatGPT began erroring around 11:00, Claude at about the same time; Anthropic attributed its outage to an "infrastructure issue" and had it resolved by ~12:15 ET. The Verge: "it's not clear what went wrong at all three companies, or if the issues were somehow related" — a shared Azure cause circulated and is unconfirmed by any of the three. (The Verge, Axios)
- Pentagon Under Secretary of War Emil Michael said Anthropic "is still a designated Supply Chain Risk" and remains one "for the Defense Industrial Base" — one day after Commerce Secretary Lutnick said the administration "trust[s] Anthropic," and a week after a federal judge permanently enjoined the designation. Bloomberg's headline: the Pentagon says its ban is on despite Lutnick. Michael's statement does not say which of the two designations this map distinguishes he means — the vacated 10 U.S.C. §3252 designation or the separate FASCSA one still pending at the D.C. Circuit — so it is either a lawful reassertion of the surviving one or a public restatement of the one a court struck down. ✏️ And the court record this map carried was wrong a second time: the Order of Final Relief (Dkt. 251), read today from the RECAP PDF, contains no stay of any kind; the injunction has run since 08-27, no notice of appeal exists, and the "7-day stay expiring 09-03" this map logged on 09-02 — and built a dated expectation on — never existed. That expectation is withdrawn. (Axios, Washington Examiner, Bloomberg, Order of Final Relief, Dkt. 251)
- Four frontier releases in three days, and each lab's most capable cyber variant is gated by the lab — none by a government. OpenAI's Daybreak (Astra), Anthropic's Cyber Verification and Life Sciences Verification programs (Mythos 5.1, the latter "in partnership with the US government"), Google's new Fairwind Program (Gemini 3.8 Flash Cyber). Neither the announcements nor the system cards name CAISI, the UK AI Security Institute, or the EO 14409 30-day pre-release framework this thread was opened to watch land on a real model. Anthropic classified Mythos 5.1 on its own CB-1/CB-2 scale; OpenAI on its own Preparedness levels. (OpenAI, Anthropic system card, Google)
- Anthropic is set to finalize a revolving credit facility at $15 billion — up from the ">$10 billion" target this map logged on 08-18, and six times the $2.5bn facility it closed a year ago. Morgan Stanley leads, with Goldman Sachs, JPMorgan and Citigroup prominent and a syndicate running to Barclays, Wells Fargo, Bank of America, Deutsche Bank, RBC, UBS, Bank of Montreal, BNP Paribas, Crédit Agricole, MUFG, Sumitomo Mitsui and Toronto-Dominion. The four leads are the same four reported to be leading the IPO. Companies typically finalize the revolver before formally notifying banks of IPO roles — SpaceX expanded a $5bn revolver a month before its own record listing — so the step matters for sequencing, not just for the size. (Bloomberg via Yahoo Finance)
- The Anthropic v. Department of War docket is unchanged. Read directly via curl across both pages: it still ends at entry #252 (Judgment, "Civil Case Terminated," 08-27). No notice of appeal has been filed. This confirms rather than changes the record — and it is worth stating plainly, because this map made two consecutive errors on this case by reading coverage instead of the docket. The appeal window runs to 09-28. (CourtListener docket 72379655)
- Anthropic's outside-trustee governance is getting its first real scrutiny as it approaches an IPO that could value it near $2 trillion. The Long-Term Benefit Trust holds no equity but can appoint or dismiss a majority of the board, and has already seated four of seven directors including Reed Hastings and Novartis CEO Vas Narasimhan. Three of five seats are filled — chair Neil "Buddy" Shah, former Fed chair Ben Bernanke, and Richard Fontaine of the Center for a New American Security. Trustees get advance notice of major actions including model launches, meet weekly among themselves and roughly biweekly with leadership, and have weighed in on real decisions including the limited Mythos rollout via the Glasswing Project. But the reporting says the trust "has not attempted to draw red lines or force a significant trade-off between profit and purpose" — so the mechanism has never been tested against a real conflict. Harvard Law's Jesse Fried calls it a "built-in conflict": investors fund a for-profit while self-appointed trustees decide how much profit to sacrifice. The trust can be removed by an 85% supermajority of voting power, a threshold that could shift once the company is public. (Ars Technica, syndicating the Financial Times)
- Anthropic's public IPO prospectus is no longer expected until late September, with marketing from mid-October at the earliest and the listing timed to close days before the November midterms — a slip from the "as early as the week after Labor Day" timing reported earlier on 09-04. Reuters (Echo Wang, 18:41 ET) also reports a $15 billion revolving credit facility being finalized before the banker-led analyst-meeting phase; sourcing is anonymous throughout, Anthropic and the four lead banks declined comment, and the plans are described as "subject to change." A direct EDGAR full-text search on the morning of 09-05 finds no S-1 or S-1/A for Anthropic. (CNBC, carrying Reuters, Investing.com, carrying Reuters)
- Anthropic said on Friday that Claude produced a complete, computer- checked formalization of Fermat's Last Theorem in the Lean proof language, working largely autonomously over eleven days: 13 million lines of Lean — more than five times the size of Mathlib, the community library the proof builds on — with 30,300 theorems proved along the way and 29,500 used in the final proof, using only Lean's standard axioms, the largest Lean proof ever written. The work ran on Prove2Me, the open formalization platform built by Tianyi Peng's group at Columbia, after a first attempt without it failed; human input was limited to occasional high-level nudges. It closes a community effort Kevin Buzzard kicked off in 2024 to formalize Wiles' proof, and it is the second lab pure-mathematics claim of the week after OpenAI's bounded-prime-gaps paper in the Astra launch materials (09-03, now in 2026-09-03-frontier-ai.md). The announcement sat in this map's own buffer six times across two days — Anthropic's post, SiliconANGLE, New Scientist among them — and was caught only by the 09-06 coverage critic, two days late. Dated 09-04; no thread holds a lab research milestone, so this is a digest-only record and a thread candidate. (Anthropic, SiliconANGLE, Anthropic on X)
- Anthropic's public IPO prospectus is no longer expected until late September, with marketing from mid-October at the earliest and the listing timed to close days before the November midterms, per Reuters on Friday evening — a slip from the post-Labor Day timing reported that morning — and the company is finalizing a $15 billion revolving credit facility before the analyst-meeting phase. Anonymous sourcing; the four lead banks declined comment; no S-1 on EDGAR as of 09-05 morning. (CNBC, carrying Reuters)
- The Anthropic v. Department of War docket still ends at entry #252, while the OpenAI copyright MDL ran from #1742 to at least #1892 by Saturday afternoon — the new entries are the public redacted versions of the Books Plaintiffs' summary-judgment papers (David Baldacci's memorandum of law and Rule 56.1 statement) and sealed supporting declarations, dated 09-05; no notice of appeal on the first (window to 09-28), no hearing date on the second, where the 09-04 entries added two more oral-argument requests and two sealed filings. The Nippon Life v. OpenAI docket still ends at the 08-04 minute entry that reset the status hearing to 09-02 — a hearing this map is tracking left no trace, and the ledger item is due 09-11. (Anthropic v. DoW docket, OpenAI MDL docket)
- The coordinated "Concord II" music-copyright docket against Anthropic moved into phased discovery at a 09-02 status conference, with a proposed case-management order due 09-23 — and its party list names Dario Amodei and Benjamin Mann personally, so the personal-liability theory that distinguished the Sony/Warner Chappell suit filed 08-28 is not unique to that case. Read from the docket; no outlet covered the conference. Dated 09-02, on the record now. (CourtListener docket 72199828)
- AI companies have pledged about $265 million to super PACs and political groups for the 2026 midterms, a Wall Street Journal analysis found — making AI, alongside crypto and betting, the leading industry spender of the cycle — The Independent reported on Saturday, framing the money as arriving "as data center backlash sweeps through communities." The parts were on the record separately: Leading the Future, the Andreessen Horowitz/Greg Brockman-backed super PAC, at $140 million raised per Reuters (a16z's $50 million and the Brockmans' $25 million in FEC filings); Anthropic's self-disclosed $40 million to Public First Action, a 501(c)(4) backing AI regulation; and Leading the Future's "Build American AI" affiliate running buildout-defence ads in Wisconsin, Ohio and Kansas since late August. New is the aggregate and its framing as a response to local opposition — the industry's political spend set against the backlash this thread tracks for the first time. A tally of pledges, not disbursements; the Journal's own analysis could not be read directly. (The Independent, via Yahoo News, WSJ, 07-22)
- The week's two frontier labs each published a pure-mathematics result alongside a product launch: Anthropic's eleven-day Lean formalization of Fermat's Last Theorem (13 million lines, 30,300 theorems proved, 29,500 used, on Columbia's Prove2Me platform) and OpenAI's "Improved Short Gaps Between Primes" (a bound of 186, Lean-formalized, "the proof is due to GPT 6 Astra") in the Astra launch materials. The two claims are of different kinds — Anthropic's is a formalization of a known theorem, checkable by running Lean; OpenAI's is a new bound, checkable by mathematicians reading a 39-page paper — and both are the sort of thing the Astra benchmark dispute this map already tracks was about: capability claims whose verification is left to outsiders. Each is curated in full on its own day; no thread holds either, and the two together are the candidate below. (Anthropic, OpenAI — paper PDF, SiliconANGLE)
- Artificial Analysis published an interim version 4.2 of its Intelligence Index on Friday 09-04 — adding a private agentic knowledge-work set (AA-Briefcase) and Surge AI's GDP.pdf, dropping the saturated GPQA Diamond, and doubling the weight on held-out sets to 40% — on which Claude Fable 5.1 leads and GPT-6 Astra is second with a four-point gain over GPT-5.6 Sol, having scored level with Sol on v4.1. AA said it had "deliberately held back updates to keep the Index stable through recent major model launches" but that "the frontier moving so quickly in the past weeks" made an interim release necessary ahead of a v5 eight months in the making; Meta is the third-ranked lab, ahead of SpaceXAI, Moonshot, Z.AI and Google. AA's note does not mention Fortune's report; The Decoder's reading that the overhaul came "likely in response to criticism" of the Astra scoring is The Decoder's. For the benchmark dispute this thread carries, the outside index whose flat score fed the skepticism now shows a gain — on a different test set, which is the point. Read in the afternoon extend; on The Enterprise Agent Land Grab under its 09-04 block. (Artificial Analysis, The Decoder)
Projects & threads
Full item history for Anthropic →