# Anthropic's War on Opensource AI > [!question] Lingering Questions > - Is the "Fable" silent-degradation claim reported or demonstrated? The essay's load-bearing anecdote rests on summarized reporting the reader can't independently verify. > - How do SB 53 / RAISE Act exemptions actually scale with startup revenue/compute — does the compliance-cost-favors-incumbents mechanism hold up against the actual exemption structure? > - What's the current measured benchmark gap between the best open-weight and best closed models, to test whether "Chinese model" risk-rhetoric is empirically justified or mostly xenophobic framing? > [!tip] Article Ideas > - A neutral, sourced timeline of every reported frontier-lab access cutoff (OpenAI, Windsurf, xAI) with primary citations — separates the verifiable record from the essay's polemical framing. > - A "model dependency SBOM" scorecard for OSPO/FAIR use: output-use freedom, access stability, retention terms, refusal transparency. > [!link] Cross Reference > - [[A Frontier Lab Can Learn From the World While Restricting the World From Learning From It]] — existing permanent note already captures this essay's central asymmetry claim (P1). > - [[Safety Regulation Advantages Incumbents When It Targets the Open Alternative]] — existing permanent note (different source) captures the same underlying mechanism as this essay's P2 (regulatory-capture-by-template); connect rather than duplicate. > - FAIR anti-capture covenants / [[3. Resources/MOCs/Open Source Governance MOC]] — this essay is essentially the FAIR thesis applied one layer down, at the model. > [!quote] Vocabulary > - **RSP (Responsible Scaling Policy)** — Anthropic's own frontier-safety framework, which the company states influenced OpenAI, Google DeepMind, SB 53, the RAISE Act, and EU AI Act codes of practice. > - **Open-weight vs. "open-ish"** — the essay's distinction between models whose weights are genuinely redistributable/fine-tunable and licenses that market openness while restricting competitive use. ## Bibliographic capsule Long-form X/Twitter essay by Ahmad (@TheAhmadOsman), ~6,700 words, posted 2026-06-12. Argues Anthropic converts safety/security framing into anti-competitive control over who may build competing AI, while conceding real abuse risks (CBRN, distillation fraud, weight theft) are genuine. ## Author's thesis (as argued) Anthropic stacks moral-authority branding, genuine frontier capability, explicit anti-competitive output/access restrictions, and policy ambition — and the combination turns "safety" into a competitive moat. The charge is structural (an institutional pattern), not an accusation of individual bad intent, and the essay's own steelman section concedes the underlying abuse risks are real. ## Key arguments (≤3) 1. **The output-use asymmetry is documented policy, not interpretation.** Anthropic's commercial terms bar using Claude outputs to train or develop competing models; Anthropic itself trained on civilization-scale data (including material from a reported $1.5B copyright settlement). 2. **Reported access cutoffs show the conflict-of-interest mechanism in action.** Anthropic revoked OpenAI's API access ahead of GPT-5 (Aug 2025) and restricted Windsurf around OpenAI's acquisition plans, citing competitive-use terms — a firm that both supplies a key input and competes with its customers can degrade or deny rivals. 3. **Anthropic's own RSP update claims to have shaped other labs' rules and multiple laws (SB 53, RAISE Act, EU AI Act codes)** — self-described influence that the essay treats as evidence the safety-regulation template is incumbent-authored, favoring capital-rich labs that can absorb compliance overhead. ## Claim atoms ### Incumbents learn from the world; the world can't learn back > [!quote] ID:rw_h1 > "The asymmetry is the whole story: Anthropic can learn from the world. The world cannot freely learn from Anthropic." Where/why: "The Anti-Opensource Rule Is Already Written Down" section — already promoted to [[A Frontier Lab Can Learn From the World While Restricting the World From Learning From It]]. ### A provider revoked a rival's API citing competitive-use terms > [!quote] ID:rw_h2 > "In August 2025, Anthropic revoked OpenAI's API access ahead of GPT-5 and pointed to terms barring use of Claude to build competing products, train competing models, or reverse engineer services." Where/why: "Customer, Competitor, Captive" section — the strongest single verifiable fact supporting the access-can-disappear thesis. ### A lab's safety policy shaped other labs' rules and laws > [!quote] ID:rw_h3 > "Anthropic's own Responsible Scaling Policy update says the RSP influenced OpenAI, Google DeepMind, California SB 53, the New York RAISE Act, and EU AI Act codes of practice." Where/why: "Why Anthropic Is Uniquely Dangerous" section — Anthropic's own statement; basis for the regulatory-capture-by-template argument, already conceptually captured by [[Safety Regulation Advantages Incumbents When It Targets the Open Alternative]]. ### "Responsible access" is most suspect from the access-seller > [!quote] ID:rw_h4 > "The most dangerous phrase in AI policy is 'responsible access' when it comes from the company selling the access." Where/why: "The Permanent Underclass Thesis" section — aphoristic compression of the conflict-of-interest argument. ### Dismissing open models as "Chinese" is a loyalty test > [!quote] ID:rw_h5 > "But dismissing open models as 'Chinese models' is intellectually lazy and politically useful to incumbents." Where/why: "Chinese Models and the Xenophobia Trap" section — the essay's own most defensible passage: judge artifacts by properties, not provenance. ## Permanent candidates - **[[A Frontier Lab Can Learn From the World While Restricting the World From Learning From It]]** — already exists (created 2026-06-28), capturing H1. Not duplicating. - **[[Safety Regulation Advantages Incumbents When It Targets the Open Alternative]]** — already exists (different source, same underlying mechanism as H3). Connect, not duplicate. - H5 ("Chinese models" as a loyalty test / xenophobia trap) is a genuine, distinct claim not yet captured — flagged as a candidate for a future pass rather than promoted today; it needs a second corroborating source before standing alone as a Permanent note. ## Processing checklist - [x] Checked for existing Permanent notes before creating — found two, linked both, created none new (claim-level duplication avoided per skill guidance). - [ ] Verify the Fable "silent degradation" claim against a primary source before citing it as settled. - [ ] Pull current open-weight vs. closed benchmark data (Stanford HAI) to test the "Chinese model" framing empirically. - [x] Connect into [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] and [[3. Resources/MOCs/Open Source Governance MOC]]. ## Metadata - Source: [[3. Resources/Readwise/Tweets/Anthropic's War on Opensource AI]] ## Links - [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] — output-use restrictions and access cutoffs are a vendor-dependency-cost question. - [[3. Resources/MOCs/Open Source Governance MOC]] — the essay's core argument is an anti-capture governance critique applied to AI models.