Safety, dignity and interoperability by design

A feminist framework for AI governance

AUDRi proposes three mutually reinforcing principles for an AI ecosystem that protects human rights, distributes technological power more fairly and places people and communities at the centre of governance.

• Safety by design: AI systems should be designed to prevent foreseeable human rights harms before they occur, rather than relying on affected communities to demonstrate harm afterwards. Safety requires participation, precaution, independent assessment and ongoing accountability, with the experiences of those most likely to be affected informing how risks are identified and addressed.
• Dignity by design: AI governance should protect people’s autonomy, agency, equality, privacy and bodily integrity. People should know when AI is being used, have meaningful opportunities to influence and challenge decisions and have access to effective, accessible and gender-responsive remedies when their rights are violated.
• Interoperability by design: AI systems and digital infrastructure should remain contestable, replaceable and accountable rather than creating dependence on a small number of proprietary providers. Shared standards should enable portability and meaningful choice while protecting privacy and preventing indiscriminate data sharing or expanded surveillance.

From principles to action

To translate these principles into practice, AUDRi calls for an independent global mechanism for AI governance with three key areas of action:

• Recognise and address exploitative digital practices: Establish or designate an independent mechanism capable of recognising, documenting and addressing practices that deepen digital power asymmetries and expose women, girls and marginalised communities to TFGBV.
• Require human rights impact and risk assessments: Governments should mandate independent pre-deployment human rights impact and risk assessments for high-risk AI systems, incorporating gendered and intersectional analysis. Where there is a credible risk of serious or irreversible harm that cannot be adequately mitigated, deployment should not proceed.
• Ensure accountability throughout the AI lifecycle: Governments should require periodic independent post-deployment assessments, particularly when AI systems are substantially modified, or new harms emerge. Companies and public bodies should disclose significant risks, incidents, safeguards, complaints and remedial measures, with high-risk systems remaining subject to regulatory oversight throughout their lifecycle.