Acme ↔ Anthropic

• high relevance·OEM
  OEM with Third-Party AI Model

  This is the canonical OEM pattern: Acme embeds Claude into a consumer product it commercializes, with Anthropic as the upstream model provider. Acme does not train the model, controls only the application layer, and is exposed to all five kill-list moves in the pattern. The consumer-facing nature of the app amplifies output-harm liability exposure compared to an internal-tool deployment.

• low relevance·Strategic investment
  Hyperscaler-Exclusive-Then-Multi-Home

  Not a hyperscaler deal, but the exclusivity dynamics are partially relevant. If Anthropic offers Acme pricing concessions in exchange for model exclusivity, or if the contract's scope language is broad enough to foreclose switching, Acme could find itself in a structurally similar position to a lab locked into a single compute provider: dependent, with no defined off-ramp.

Failure modes from the matched patterns that are specifically relevant to this deal's structure. Each one comes with the pattern's mitigation excerpt where applicable.

1. 1.
   Asymmetry-without-protection

   from: OEM with Third-Party AI Model →

   Acme is a smaller enterprise building a consumer product on top of a frontier lab's model. Anthropic controls model availability, pricing, usage policies, and deprecation timelines. If the contract does not include advance notice periods for material changes, migration rights, and SLA commitments, Acme's product roadmap is hostage to Anthropic's unilateral decisions. This is the structural center of gravity for this deal.

   Mitigation. Negotiate minimum advance notice (e.g., 12 months) for model deprecation or material pricing changes, with a migration window and rate-lock.

2. 2.
   Indemnification-follows-the-contract-chain

   from: OEM with Third-Party AI Model →

   Because this is a consumer-facing app, end users harmed by a generated business model (e.g., legally defective advice, hallucinated financial projections) may sue Acme. Without explicit multi-party liability allocation, Acme cannot assume Anthropic's indemnification flows through automatically. Acme could absorb costs for failures originating in the model layer.

   Mitigation. Require explicit indemnification carve-outs covering model-layer failures versus application-layer failures, with mutual liability caps tied to contract value.

3. 3.
   Deferred IP allocation

   from: OEM with Third-Party AI Model →

   If Acme fine-tunes Claude on proprietary datasets (e.g., user-generated business model inputs, curated templates), the ownership of those fine-tuned weights and the derivative training data needs to be resolved now, not later. Anthropic's standard commercial terms may claim rights over fine-tuning contributions that Acme considers core product IP.

   Mitigation. Define at signing who owns fine-tuned model weights, what Anthropic may do with Acme's fine-tuning data, and what happens to those assets on termination.

4. 4.
   Premature exclusivity

   from: OEM with Third-Party AI Model →

   If Anthropic offers favorable API pricing contingent on Acme committing to Claude as its sole model provider, Acme loses the ability to multi-home (e.g., add GPT-4o for cost optimization or redundancy) without paying a penalty. For a consumer app at early scale, locking in before understanding unit economics is a structural mistake.

   Mitigation. Resist exclusivity tied to pricing tiers. If exclusivity is accepted, define a sunset trigger tied to volume thresholds or contract anniversary dates.

• Anthropic ↔ Salesforce
  Claude becomes preferred Agentforce model for regulated industries with deep Slack integration

  Anthropic-Salesforce is also an embedded-Claude arrangement where a third-party platform (Salesforce's Agentforce) surfaces Claude to end users. The ISV embedding dynamic and the downstream user liability questions are structurally analogous to Acme's consumer app.

• Anthropic ↔ Databricks
  Five-year deal: Claude natively available on Databricks Data Intelligence Platform

  Anthropic-Databricks is a model-hosting OEM arrangement where Claude is natively embedded in a commercial platform. The IP allocation and output-rights questions that arise when a third-party platform intermediates Claude to end users are directly relevant to Acme's situation.

• Deloitte ↔ Anthropic
  Deloitte deploys Claude to 470,000 employees globally, Anthropic largest enterprise deal

  Deloitte's large-scale Claude deployment illustrates how enterprises handle Anthropic's enterprise agreement structure at scale, including usage policy constraints and indemnification. Acme is smaller but faces the same contractual architecture from the weaker bargaining position.

From the matched patterns. Each is hypothetical and illustrative; the full clause text with annotations is on the linked pattern page.

• Exit triggers and migration rights (termination)
  from: OEM with Third-Party AI Model →

  This is the highest-priority clause for Acme. A consumer product cannot absorb a sudden loss of its core AI capability. The clause should define minimum notice periods for termination or material change, a post-termination wind-down period with continued API access, and Acme's right to export any fine-tuned assets.

• Liability allocation for output harm (liability_indemnification)
  from: OEM with Third-Party AI Model →

  Consumer-facing business model generation creates real downstream harm exposure. This clause should draw a clean line between model-layer failures Anthropic is responsible for and application-layer failures Acme owns, with explicit indemnification scope and mutual liability caps.

• Fine-tuning IP allocation (ip_allocation)
  from: OEM with Third-Party AI Model →

  If Acme customizes Claude using proprietary data, the ownership of resulting weights and the permissible uses of Acme's training data by Anthropic must be explicit at signing. Leaving this to standard API terms is insufficient if fine-tuning is part of Acme's product differentiation strategy.

• Output ownership and use rights (license_grant)
  from: OEM with Third-Party AI Model →

  Acme needs a clear grant of rights to commercialize Claude-generated outputs to its consumers. The license grant should confirm that outputs are Acme's to use, sublicense to end users, and incorporate into downstream deliverables without restriction beyond usage policy compliance.

• Sunset exclusivity with defined trigger (scope_limits)
  from: Hyperscaler-Exclusive-Then-Multi-Home →

  If any exclusivity is accepted in exchange for pricing concessions, this clause ensures Acme has a defined path off that exclusivity tied to objective triggers (volume, time, or pricing change), rather than leaving restructure to ad-hoc negotiation when Acme is in a weaker position.

• ?Does Anthropic's commercial API agreement give Acme adequate advance notice and a rate-lock period before pricing changes or model version deprecations, and if not, is Acme willing to accept that product continuity risk?
• ?Will Acme fine-tune Claude on user inputs or proprietary datasets, and if so, has Acme reviewed Anthropic's terms to confirm it retains ownership of those fine-tuned assets on termination?
• ?How is liability allocated when a consumer suffers harm (financial, legal, reputational) from acting on a business model the app generated, and does Acme's current indemnification posture reflect that it cannot pass that liability upstream to Anthropic under standard terms?
• ?What is Acme's model substitution or multi-model fallback plan if Anthropic raises prices materially, restricts usage policy in ways that affect Acme's use case, or experiences a sustained outage?
• ?Does the app's consumer-facing nature trigger any regulatory obligations (e.g., FTC guidance on AI-generated financial or business advice, EU AI Act transparency requirements) that the contract with Anthropic does not address, and who bears compliance cost?