Turing is the one confirmed direct buyer — run a 10-target bid process around it
The question: beyond HUD.so’s approach, who else should we invite to bid on licensing our private codebase for AI training? This report ranks 10 prospects (research as of 2026-08-30) on two 1–5 scores: strength of public evidence that the route buys code, and informed bid potential. Bid potential is a prioritization score, not a quoted offer.
Bid-target matrix: evidence vs bid potential
Route mix
How the 10 targets break down by route type, with the leading prospect on each route.
| Route type | Prospects | Avg bid potential /5 | Best /5 | Leading prospect | Source |
|---|
What the matrix cannot say alone
Turing / Project Lazarus is the only prospect scoring 5/5 on both evidence and bid potential — reported to acquire failed-startup and legacy production code for model training.
webpronews.comGoogle’s reported pilot pays selected Play Store developers for private Android app source — invitation-led, and strongest if the asset is an Android app.
9to5google.comPangea and FileYield are intermediaries: their quotes are not independent end-buyer bids — demand named buyers, commissions and deletion terms before sharing code.
fileyield.comMistral, Cognition and Augment Code are strategic prospects with code-model relevance, but none has a publicly confirmed code-purchase program — treat them as cold, targeted proposals.
mistral.ai
The shortlist, ranked
| # | Prospect | Route | Ev/5 | Bid/5 | Rationale & approach | Caveat | Source |
|---|
Outreach playbook
- Prepare a two-page data-room teaser before granting repo access: languages and frameworks, lines of code, commit count and date span, git history, tests, issues and PRs, docs, deployment artifacts, domain complexity, provenance, contributor and IP ownership, third-party and open-source dependencies.
- Solicit written, comparable terms on a common deadline from all 10 targets.
- Ask each bidder to state: license scope (training, fine-tuning, evaluation, embeddings, derived and synthetic data), named affiliates and customers, exclusivity, term and territory, sublicensing, retention and deletion, model-weight survival after termination, security and audit, indemnity, attribution, payment timing, and representations.
- Require an NDA and staged access: metadata first, sample second, full repository only in a secure clean room or controlled transfer.
- Preserve leverage: prefer non-exclusive rights unless the exclusivity premium is substantial.
Legal and security cautions
- Verify the company owns all code and training-data rights: inspect employee and contractor assignments, customer code, trade secrets, credentials, PII, telemetry, third-party code, copyleft obligations, patents, export controls, and security-sensitive material.
- Do not grant HUD or any intermediary exclusivity, a right of first refusal, broad agency authority, or buyer-confidentiality terms that block bid comparison before counsel reviews them.
- Distinguish intermediary indications from independent end-buyer bids, and identify commissions.
- This is commercial research, not legal advice — engage specialist IP and data-licensing counsel.
Live web research as of 2026-08-30 across diverse searches and primary or credible secondary pages; 10 ranked prospects plus an 8-row route-type summary. Evidence and bid-potential scores are 1–5 analyst ratings; bid potential is a prioritization score, not a quoted offer. Direct evidence is strongest for Turing’s reported acquisitions, OpenAI’s public form, OpenDataBay’s code marketplace, FileYield’s brokering claims and Mistral’s private-repo model work; Google appears selective and invitation-led; Pangea’s indexed materials are thin and need enhanced diligence. Full approach routes were shortened in the table for space.