Built for privileged work.
Firms run client matters through Patent Bear. Here is how the service is designed to support confidentiality and attorney work-product protection, and where the honest limits are.
Last reviewed: September 29, 2026
- 01
A contractual commitment, not a policy statement
Your materials are your confidential informationOur Terms of Service say it plainly: Patent Bear treats the materials you submit, including invention disclosures, claim sets, product descriptions, prompts, and uploaded documents, as your confidential information. We use them solely to provide the Service to you, we do not sell them, and we do not disclose them to anyone other than the subprocessors listed in our Privacy Policy as needed to provide the Service.Zero Data Retention with every AI providerEvery AI call Patent Bear makes runs under Zero Data Retention, in writing. OpenAI has enabled Zero Data Retention on our organization under a signed Modified Retention Amendment, and our Data Processing Agreement with DeepInfra commits it to Zero Data Retention as well. When your prompts, claims, and work product go to an AI provider, the provider processes them to answer the request and keeps nothing: not logged, not kept for abuse monitoring, not available for human review, and never used for training. What you save stays in your Patent Bear account.No training on your data, ours or our providers'Patent Bear does not train AI models on your submissions or outputs. Our AI providers, OpenAI and DeepInfra, are barred under their API terms from training on customer inputs and outputs unless the customer opts in. Patent Bear does not opt in. We use only their business API tiers, never consumer products like ChatGPT, so consumer-tier data practices never apply to your work. - 02
A deliberately small disclosure surface
Two AI providers, publicly listedFor Patent Bear's own processing, exactly two AI providers receive content from your work: OpenAI (response generation) and DeepInfra (passage ranking). The full subprocessor list is published in our Privacy Policy, the vendor transparency a conflicts or vetting review asks for.Your assistant, your agreementIf you use Patent Bear through a connected AI assistant such as Claude, ChatGPT, or Gemini, the results you request are returned to that assistant, and its provider handles them under your firm's agreement with that provider, not ours. Choose an assistant plan whose terms fit client work, and disconnect any time from your plan page.Server-side onlyAll AI calls originate from Patent Bear's servers over encrypted connections. Your browser never talks to an AI provider directly, and no browser extension or third-party script sees your content.Stateless endpoints, no provider-side storageWe call only stateless API endpoints: no provider-side assistants, file stores, or conversation storage, so there is nothing for a provider to hold between requests. With Zero Data Retention in place, the provider keeps nothing after the response either. The one exception is images of published patent drawings, which are public documents and fall outside OpenAI's retention amendment. Your durable data lives only in Patent Bear's own systems.Single-tenant workers, anonymous job keysResearch jobs run on Patent Bear's own private, single-tenant worker services, not a shared pool. Jobs are keyed by an anonymous account identifier and processed transiently in memory; work product is not persisted to worker disk. - 03
Isolation and access control
Row-level security at the databaseEvery query is scoped to the authenticated user by row-level security policies enforced at the database itself, not just in application code. Encryption is AES-256 at rest and TLS 1.2+ in transit; full detail is on our Security page.No third-party trackingNo advertising pixels or cross-site tracking. Analytics, when enabled, is cookieless and aggregate, and never receives raw queries, pasted work product, credentials, or billing details. - 04
Work-product framing
Attorney Work Product legends on exportsEvery Research report can be exported with an Attorney Work Product legend, so documents generated in anticipation of litigation carry the framing your file expects from day one.Matter organizationRuns, drafts, conversations, and saved documents can be organized under matters, keeping client work separated and easy to produce, transfer, or destroy on a per-matter basis. - 05
Deletion you can actually exercise
Per-run and per-conversation deletionDelete individual Research runs, drafts, conversations, and uploaded documents at any time. Deletion removes the underlying records and stored files, not just the listing.Account deletionRequest account deletion and we remove your active application data, including annotations, saved patents, runs, drafts, uploads, and account records, subject only to limited backup retention. - 06
What we won't claim
Privilege is a legal conclusion, not a featureWhether attorney-client privilege or work-product protection attaches to any particular use depends on your facts, your jurisdiction, and how your engagement is structured. Patent Bear is built to support the confidentiality posture those doctrines require: contractual confidentiality, no training, minimal and disclosed subprocessors, transient processing, and deletion rights. But no vendor can promise privilege, and we won't pretend otherwise.Vet usWe welcome security questionnaires and vendor diligence. If your firm needs documentation for an AI-use or outside-vendor review, write to us and we'll work through it with you.