Auditable Commercial License v1.0

A new Auditable Commercial License v1.0 (ACL 1.0) has been released, designed to address AI training data protection—a gap in earlier commercial source-available licenses like BUSL and Elastic 2.0. The license prohibits using licensed work as AI training data, includes a safe harbor for ordinary AI-tool use, and automatically converts to Apache 2.0 after four years. It targets enterprise buyers and regulated industries needing explicit audit rights and AI-era protections.

Read, run, and modify the source internally. No redistribution, no hosted service. Every version becomes Apache 2.0 in four years — with the AI-training protection older commercial licenses were never built to address. Licensees may read, review, and audit the source code for security, compliance, and operational purposes — the rights enterprise buyers and regulated industries actually need, written in explicitly rather than implied. Section 2.3 prohibits use of the Licensed Work as AI training data, with a knowledge qualifier, a safe harbor for ordinary AI-tool use, and a flow-through obligation to AI vendors. The first commercially drafted clause built for the agent era. Every released version automatically converts to the Apache License, Version 2.0 on the fourth anniversary of its public release. Adopters are never permanently dependent on the licensor's goodwill for any given version. | Apache 2.0 | BUSL 1.1 | Elastic 2.0 | PolyForm | ACL 1.0 | | |---|---|---|---|---|---| | Source available | Yes | Yes | Yes | Yes | Yes | | Blocks hosted-service competition | No | Yes | Yes | Partial | Yes | | AI training protection | No | No | No | No | Yes | | Auto-converts to OSI license | N/A | After 4y | No | No | After 4y | | Tier structure built in | No | No | No | No | Yes | | Enterprise IP indemnification | No | No | No | No | Yes | | Contributor patent grant | Yes | Partial | Partial | No | Yes | LICENSE.md — drop it at the root of your repository. Commercial source-available licenses were drafted for a world that didn't have autonomous code assistants. BUSL in 2017. Elastic 2.0 in 2021. None addresses the threat that now matters most — that the act of working with source code, using the tools developers actually use, quietly becomes an act of exfiltration into someone else's model. This license addresses it. Three layers: a knowledge-qualified prohibition, a safe harbor for ordinary AI-tool use, and a flow-through obligation that reaches the vendor relationship without requiring surveillance. The clause is not proprietary. Adopt it, adapt it, improve it. What matters is that the category has one — because every commercial license drafted in 2026 and 2027 will eventually need it, and the version that becomes standard will be the version written with the enforcement problem in mind.