[
  {
    "market": "embedded-evals",
    "source": "Anthropic commitment",
    "url": "https://darioamodei.com/post/we-must-pace-the-frontier",
    "use": "Immediate intention to establish continuing employee-like access and independent reporting; supports preparation advantage."
  },
  {
    "market": "embedded-evals",
    "source": "OpenAI endorsement",
    "url": "https://x.com/sama/status/2098811563415150910",
    "use": "Same-day commitment; originator/follower distinction concerns preparation, not a long elapsed announcement lead."
  },
  {
    "market": "embedded-evals",
    "source": "METR at Anthropic",
    "url": "https://metr.org/blog/2026-03-25-red-teaming-anthropic-agent-monitoring/",
    "use": "Three-week internal exercise with system access and staff cooperation; evidence of operational feasibility. Onboarding duration unknown."
  },
  {
    "market": "embedded-evals",
    "source": "METR at OpenAI",
    "url": "https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/",
    "use": "Investigation began July 29 evening; report August 26. The 28-day interval includes investigation and publication; excluded from onboarding timing."
  },
  {
    "market": "embedded-evals",
    "source": "Delmas and Montes-Sancho (2010)",
    "url": "https://www.ioes.ucla.edu/wp-content/uploads/2017/04/2010-Delmas-Montes-Sancho-Strategic-Management-Journal.pdf",
    "use": "Direction of originator/preparation advantage. 62 early, 21 late, 49 nonparticipants; delayed entry differs from a same-day follower. No transferable duration coefficient."
  },
  {
    "market": "embedded-evals",
    "source": "Toffel (2006)",
    "url": "https://www.hbs.edu/ris/Publication%20Files/07-023_1c3822d6-d853-47f6-be07-c10a97ee4218.pdf",
    "use": "Early adopters showed stronger subsequent improvements; shorter follow-up for later adopters limits interpretation. Directional support only."
  },
  {
    "market": "external-first",
    "source": "OpenAI GPT-5.3-Codex",
    "url": "https://openai.com/index/introducing-gpt-5-3-codex/",
    "use": "Developer reports using early model versions in its development; evidence of benefits from early internal use."
  },
  {
    "market": "external-first",
    "source": "IAPS internal-use risk reporting",
    "url": "https://www.iaps.ai/research/risk-reporting-for-developers-internal-ai-model-use",
    "use": "Internal deployment creates a separate governance problem; motivates potential pressure for restrictions."
  },
  {
    "market": "external-first",
    "source": "Anthropic pacing proposal",
    "url": "https://darioamodei.com/post/we-must-pace-the-frontier",
    "use": "Discusses restricting internal use to improve AI; evidence of policy interest. This specific external-first rule still requires adoption."
  }
]