The fourth meeting of the Signatory Taskforce for the General-Purpose AI (GPAI) Code of Practice focused on the Safety and Security Chapter and the Copyright Chapter.

Regarding safety and security, the AI Office presented how analyzing model usage can support post-market monitoring and systemic risk assessments. According to the Office, information on how AI models are used can complement pre-deployment evaluations and improve the overall risk management process.

Participants also discussed the use of “marginal-risk” clauses in Safety and Security Frameworks. These provisions would allow providers, in exceptional circumstances, to consider deploying less restrictive models if competitors were to release unsafe frontier models. The AI Office stressed that such clauses should only be invoked under strict evidentiary and procedural safeguards.

On the Copyright Chapter, discussions focused on the obligations of GPAI model providers to publish information about the web crawlers they use and their robots.txt features. Signatories also committed to making public any measures adopted to identify and comply with rights reservations expressed by copyright holders, while providing an automated notification mechanism whenever this information is updated.

The meeting concluded with an overview of ongoing initiatives related to the GPAI Code of Practice and a discussion on the organisation of future Signatory Taskforce meetings.