The “Open Source AI Definition 1.0” (OSAID) recently published by the Open Source Initiative (OSI) provides clear rules in the area of “Open Source AI”. For the first time, the definition provides clear guidance as to which requirements an AI system must meet in order to be considered truly “open” – in particular, transparent, modifiable and accessible to the public. I wrote a few lines about this in my blog on software law:
Background: In the definition now presented, the OSAID requires that an open source AI be disclosed in terms of its use, functionality and modifiability as well as in its distribution. This openness requires detailed provision of data information, source code and parameters so that experts can understand and modify the models. This definition is an important step, but critical voices are already increasing.
The reason? Many large AI providers, including Meta and Elon Musk’s Grok, market their systems as “open”, but without meeting all the requirements for transparency and access. This is called “open washing” – marketing with supposed openness, while behind the scenes access remains restricted. This is reminiscent of well-known forms of “AI washing” and could distort competition in the long term. An exciting aspect here is how the definition affects copyright issues: training data protected by copyright and a lack of clarity about its use are already causing conflicts that are only likely to become more present as a result of the OSAID.
My short contribution to the “Open Source AI Definition 1.0” is aimed at everyone who is involved in the field of AI and open source or who wants to find out about the new definition and its critical assessment. It marks an important step, but also leaves questions unanswered – an exciting standard with future potential! And of legal importance, because essentially, violations in AI washing can ultimately also be prosecuted under competition law.
Last articles by lawyer Jens Ferner (specialist lawyer for IT & criminal law) (Show all)
