ACSC releases guidance on using open‍-‍weight AI models

The Australian Cyber Security Centre (ACSC) has released two new publications to help organisations make informed decisions about adopting and operating open-weight AI models.

Open-weight AI models can provide organisations with greater control, flexibility and resilience when deploying artificial intelligence capabilities. They can be hosted and managed within an organisation’s own environment, enabling greater control over data, operations and lifecycle management.

The first publication, Open-weight AI models: Opportunities and cyber security risks, explains what open-weight AI models are, the benefits they can provide, and the cybersecurity and supply chain risks that arise from the public release of model weights.

The second publication, Selecting, deploying and operating open-weight AI models securely, provides practical guidance on governance, model selection, secure deployment, monitoring, maintenance and incident response for organisations operating open-weight AI models.

Organisations that choose to deploy and operate open-weight AI models have to assume greater responsibility for governance, security, monitoring, maintenance and incident response. Effective risk management is essential throughout the model lifecycle.

Together, the publications aim to provide guidance on:

• understanding the opportunities and risks of open-weight AI models
• assessing model provenance and trustworthiness
• managing model and software supply chain risks
• establishing governance and oversight arrangements
• securing deployments and connected systems
• implementing monitoring, maintenance and incident response processes.
   

Organisations considering or operating open-weight AI models can use this guidance to better understand the risks, responsibilities and security controls required to support their safe and secure adoption.

Image credit: iStock.com/ispyfriend