SAFEnet - Southeast Asia Freedom of Expression Network

The Dialogue would be a success if: 1. The Dialogue recognized that AI governance divides exist within countries, not only between them. Capacity-building programmes should include civil society organisations, unions, affected communities, and sub-national regulators, not only national governments and industry. 2. The Dialogue called on states to establish or strengthen independent, accessible redress mechanisms for AI-related harms, with particular attention to gender-based violence, labour rights, and discrimination. Redress should not depend solely on the willingness of AI companies to self-regulate. 3. National AI governance frameworks should integrate human rights due diligence across the entire AI lifecycle, drawing on the UNGPs and the B-Tech Project's guidance. The Dialogue encouraged the sharing of practical HRDD implementation experiences across jurisdictions. 4. The Dialogue supported the exchange of regulatory sandbox experiences, particularly from the Global South, to build an evidence base for context-sensitive, interoperable governance approaches. 5. The Dialogue's multi-stakeholder format included mechanisms for affected communities (gig workers, women subjected to deepfake violence, creative workers whose output is used for AI training) to contribute testimony, not only through CSO representatives but through direct participation where feasible.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

SAFEnet selected these four areas because they correspond to the harms we document daily and the governance gaps we encounter in Indonesia's AI policy process. 1. Social, economic, ethical, cultural, linguistic and technical implications: SAFEnet's casework shows AI reshaping livelihoods and information environments right now. Ride-hailing drivers in Jakarta and Yogyakarta face algorithmic wage-setting, automated suspension, and opaque order allocation with no avenue for appeal. Generative AI tools trained on copyrighted work threaten Indonesia's creative economy. During the 2024 general election, AI-generated disinformation spread through recommendation algorithms faster than verified reporting. 2. Protection and promotion of human rights: Indonesia's Ministry of Communication and Digital reported a sharp increase in harmful deepfake cases in 2025, many targeting women and children. SAFEnet's complaint channel has received a growing number of AI-facilitated gender-based violence reports. In documented cases, police refused to act because existing laws did not cover the specific technology used. AI systems trained on Global North datasets also reproduce biases that do not map onto Indonesian social structures, affecting persons with disabilities, gender minorities, and marginalised communities. 3. Safe, secure and trustworthy AI: Indonesia is drafting a Presidential Regulation on AI Ethics, but the current text exempts "national security" uses from monitoring without defining the term. The civil society coalition recommended deleting this blanket exemption. Safety frameworks must be grounded in binding regulation, not voluntary ethics alone. 4. Transparency, accountability, and human oversight: Indonesia's AI roadmap does not establish an effective redress mechanism. SAFEnet and its coalition partners have proposed mandatory human rights due diligence across the AI lifecycle, a liability model based on reversed burden of proof under existing consumer-protection law, and a permanent multi-stakeholder forum with binding consultative authority. Without these, affected communities have no way to contest algorithmic decisions or seek remedy.

Indonesia's experience offers a concrete case study for the Dialogue. The country is the world's fourth most populous, has a fast-growing digital economy, and is building its AI governance architecture in real time. Indonesia is working through three instruments: the 2023 Ministerial Circular Letter on AI Ethics (SE Menkominfo No. 9/2023), the national AI Roadmap released on 11 August 2025, and a draft Presidential Regulation on AI Ethics under public consultation since early 2026. The roadmap was drafted by a task force of over 300 participants from industry, academia, government, and civil society, divided into seven working groups and following UNESCO's AI Readiness Assessment Methodology (AI RAM) in 2024. Indonesia is building its AI governance framework while AI-related harms are already occurring. The gap between the two is where damage concentrates. We identify particular challenges in information manipulation and AI-enabled disinformation, gendered AI harms, algorithmic labour control in the platform economy, impacts to copyright and the creative economy, environmental accountability. As part of the ongoing work, we foresee opportunities to decolonize AI ethics, incorporate human rights due diligence, adressing the AI redress gap, and building civil society's capacity to address challenges presented by the governance gaps.

1. The Dialogue should call on states to establish or strengthen independent, accessible redress mechanisms for AI-related harms, with particular attention to gender-based violence, labour rights, and discrimination. Redress should not depend solely on the willingness of AI companies to self-regulate. 2. National AI governance frameworks should integrate human rights due diligence across the entire AI lifecycle, drawing on the UNGPs and the B-Tech Project's guidance. The Dialogue should encourage the sharing of practical HRDD implementation experiences across jurisdictions. 3. The Dialogue should support the exchange of regulatory sandbox experiences, particularly from the Global South, to build an evidence base for context-sensitive, interoperable governance approaches.

The Dialogue's multi-stakeholder format should include mechanisms for affected communities (gig workers, women subjected to deepfake violence, creative workers whose output is used for AI training) to contribute testimony, not only through CSO representatives but through direct participation where feasible.