Prime Radiant/Machine Cards
DMDayCANON

Claude / Anthropic Foundation Model Lab (2021)

infrastructure pace layer · 2021–ongoing

lifespan: 50 yrs · motor: push

Anthropic is a DM-Day instance of the foundation-model-lab class archetype (class card: machine:openai-foundation-model-lab-2015). Founded December 2021 as a Delaware Public Benefit Corporation (PBC from inception — structurally distinct from OpenAI's nonprofit → capped-profit transition) by Dario Amodei (CEO), Daniela Amodei (President), Tom Brown, Chris Olah, Sam McCandlish, Jack Clark, and Jared Kaplan — all departing OpenAI over alignment-research-priority disagreements. The Anthropic-distinctive motor is push with a safety-research-first orientation: Constitutional AI (CAI) paper Dec 2022; Responsible Scaling Policy (RSP) Sept 2023; Sleeper Agents safety paper Jan 2024; mechanistic interpretability program (Chris Olah group). Commercial deployment is the flywheel that funds safety research, not the terminal goal. Claude model family: Claude 1 (limited beta March 2023); Claude 2 (public API July 2023, 100K context); Claude 2.1 (200K context, Nov 2023); Claude 3 family (Opus / Sonnet / Haiku, March 2024 — Opus benchmarked above GPT-4 on several tasks); Claude 3.5 Sonnet (June 2024, new SOTA for coding/reasoning); Claude 3.5 Sonnet new + Haiku 3.5 (Oct 2024); Claude 3.7 Sonnet (Feb 2025, extended thinking / hybrid reasoning); Claude 4 / Opus 4 / Sonnet 4 (May 2025); Opus 4.1 (Aug 2025); Sonnet 4.5 (Sept 2025); Haiku 4.5 (Oct 2025); Sonnet 4.6 + Opus 4.6 with computer-use capability. Claude Code launched Feb 2025 (agentic coding assistant; this very card is authored by a Sonnet 4.6 instance of Claude Code — meta-instance relationship to LM target machine:dawn-machine-substrate-knowledge-class). Funding: Series A $124M (2022); Google $300M+ (Feb 2023); Series C $450M (May 2023); Amazon $4B (Sept 2023) + $4B (March 2024) + $4B (Nov 2024) = $13.7B total Amazon commitment by end-2024 (AWS as primary cloud provider and co-investor; Bedrock distribution partnership); Spark Capital + Google $2B (Oct 2023); Series F $13B (March 2025; $61.5B valuation). ~$3B+ ARR December 2024 (EXTRAP; multiple press estimates). Partnerships: AWS Bedrock (distribution), Google Cloud (TPU access + investment), Salesforce, Adobe (enterprise API consumers). v0.2 GAP: Substrate.institutional NOT in SubstrateType enum — Anthropic's PBC legal form is a substrate but cannot be encoded in the substrate list; recorded here in description. [STUB-substrate-enum-gap] Commodity enum gap: API-requests-served, tokens-generated, safety-evaluations-run, interpretability-features- studied have no Smil commodity match → commodity: null + [STUB] in inputs/outputs. Intra-DM couplings: class parent OpenAI lab, competitor Google DeepMind, competitor Chinese AI labs, depends-on AWS cloud infrastructure, depends-on Microsoft Azure (Bedrock runs on AWS not Azure; distinct from OpenAI-Azure). Cross-era: parasitic_extraction → MM university (training corpus + PhD talent); zombie_dependency → MM electrical grid; LM counter-instance adapted_inheritance (capture-resistant-ai-infrastructure-class); LM meta-instance for Claude Code (dawn-machine-substrate-knowledge-class).

Machine type

incorporeal

Plasticity

plastic

Substrate

incorporeal cognitive semiotic

Wave source

wave-9-atlas

Inputs

  • aws_amazon_investment_and_compute_partnership
  • gpu_tpu_cluster_compute_via_aws_google_cloud
  • phd_researcher_alignment_and_interpretability_labor
  • pretraining_corpus_rlhf_constitutional_ai_feedback_data

Outputs

  • claude_model_family_api_and_consumer_products
  • constitutional_ai_and_safety_research_outputs
  • api_revenue_annual_recurring
  • safety_narrative_agi_risk_framing_for_industry

Landscape pressures

  • compute_arms_race_capital_intensity (85% intensity)
  • open_weights_competition_meta_deepseek_mistral (78% intensity)
  • regulatory_ai_governance_eu_ai_act_us_eo (72% intensity)
  • safety_vs_capability_internal_mission_tension (62% intensity)
  • copyright_training_data_litigation_risk (68% intensity)

Intra-era couplings

Cross-era couplings

State variables

heretic_density
0.70
CANON
ontological_doubt_prevalence
0.65
CANON
real_virtuality_saturation
0.72
CANON
opp_strength
0.72
CANON
capture_resistance_index
0.35
CANON
gravitational_weight
0.82
CANON
black_box_depth
8
CANON
push_fragmentation_count
12
CANON

Phase snapshots

DM-Dawn2021–2023complex
DM-Day2023–2024complex
DM-Day2024–2026complex

Sources

  • Anthropic (2022). Constitutional AI: Harmlessness from AI Feedback (Bai et al., 2022) · 92%
  • Anthropic (2023). Responsible Scaling Policy (RSP v1.0, September 2023) · 90%
  • Anthropic (2024). Model Card and Evaluations for Claude Models (2024) · 88%
  • Wired / The Information / Bloomberg (2025). Anthropic funding round reporting (2022–2025) · 78%
  • Sleeper Agents paper (Hubinger et al.) (2024). Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training · 90%
  • Atlas (Prime Radiant) (2026). research/09-atlas/dm-mm-industrial-stubs/findings.md (DM-AI-cluster) · 82%