Vendor drift

DigitalOcean

Classified material changes to DigitalOcean’s public trust documents, from continuous monitoring.

Data as of 2026-09-13 — refreshed on a rolling bake, not live. · digitalocean.com

Recent material changes

• high · AI feature launched · 2026-03-16 · ai_disclosure

  DigitalOcean launched a public preview of an NVIDIA Nemotron Super 120B inference offering, a hosted LLM service for agentic AI and complex reasoning. The service processes customer input data through AWS, Cloudflare, and Traversal subprocessors, and while DigitalOcean states it does not use input/output data for model training by default, customers authorize use of technical metadata and feedback for service improvement.

• medium · Training rights changed · 2026-08-08 · tos

  DigitalOcean's AI posture changed: training permission prohibits → silent.

• medium · Subprocessor changed · 2026-07-29 · subprocessor_list

  DigitalOcean added 1 subprocessor(s): Data Center Services.

• high · AI feature launched · 2026-07-20 · tos

  DigitalOcean expanded its AI Features disclosure in Section 4.8 to explicitly restrict use in high-risk scenarios, defining them as circumstances where AI failure could result in death, personal injury, severe physical/property/environmental damage, or significant legal effects. The vendor also clarified that AI Features may rely on third-party models or services and added language disclaiming liability for indirect damages and limiting aggregate liability to amounts paid for the applicable Beta Services.

• high · AI feature launched · 2026-03-12 · ai_disclosure

  DigitalOcean launched a public preview of MiniMax-M2.5 inference offering, a hosted AI inference endpoint for text-to-text generation. The service involves three subprocessors (AWS, Cloudflare, Traversal) and processes customer input/output data, with explicit language that DigitalOcean does not use data for model training unless separately opted-in, but does use technical metadata and feedback for service improvement.

• high · AI feature launched · 2026-03-31 · ai_disclosure

  DigitalOcean launched the Arcee Trinity Large Inference Offering, a new AI-powered service providing hosted inference endpoints for the Trinity Large model developed by Arcee AI. The service processes customer input data through the model and explicitly permits DigitalOcean to use technical metadata and customer feedback to improve infrastructure, with model training opt-out available only via separate agreement.

• high · AI subprocessor added · 2026-06-22 · ai_disclosure

  DigitalOcean published a new Third-Party AI Models disclosure document listing 15+ AI subprocessors including Anthropic Claude, OpenAI, Meta Llama, Mistral, DeepSeek, Google Gemini, and others. This represents a significant expansion of AI subprocessors that may process customer content through DigitalOcean's platform, with some models available only as beta services.