Contact
03 Jul 2026

Drawing the Line: Protecting Human Agency Requires AI Red Lines

After the AI Impact Summit earlier this year, we warned that AI governance is drifting into a void and pointed to growing discussions around AI red lines as a realistic way forward. 

Now as governments and stakeholders gather in Geneva next week for the inaugural UN Global Dialogue on AI Governance, Global Partners will be calling for AI red lines: clear prohibitions on AI systems and uses that pose unacceptable risks to human rights. 

—

What’s at stake? The erosion of human dignity and agency

Recent advances in generative and agentic AI have made systems that produce human-like text, images, and interactions widely accessible, yet their limitations remain poorly understood by the people using them. This is compounded by “automation bias”: the tendency to trust automated outputs as objective or authoritative. The result is not a distant risk but a present reality: fabricated facts, proliferating deepfakes, and manipulative interactions, with the most alarming cases involving the exploitation or psychological harm of young, vulnerable, or marginalised users. Agency disappears at exactly the moment people can no longer exercise control, judgment, and responsibility over the technology in front of them.

The response lies in AI red lines

In this context, the concept of “red lines” has emerged as a critical tool for governance. Red lines are clear prohibitions on AI systems and uses that pose unacceptable risk to human rights. They mark the difference between harms that can be reduced by better design or oversight and harms that are inherent to a system’s purpose and where no safeguard is sufficient. In this sense, red lines make explicit for AI a line that governance has always had to draw: between what a technology may be permitted to do, and what it must never be allowed to do. 

None of this is radical, despite often being framed this way. International human rights law already provides a robust, universally recognised framework, binding both states and companies to respect rights such as privacy, freedom of expression, and equality. And it already requires that practices which cannot be shown to be necessary, proportionate, and lawful be restricted. Red lines simply apply that same logic to AI. There is broad agreement from a broad set of actors – governments, technologists, civil society, faith leaders – on where several of them belong, including biometric mass surveillance, social scoring, predictive policing, and systems designed to manipulate behaviour or exploit vulnerabilities. What these share is a common effect: they all harm autonomy and reduce individuals to objects of control or exploitation.

Priorities for the UN Global Dialogue on AI Governance

The Dialogue represents a critical opportunity to close the AI governance accountability gap and move from consensus to implementation. As states and other stakeholders converge to shape the future of AI, here are our key messages:

  • Advances in AI are bringing about harms that erode human dignity and agency. Systems that enable mass surveillance, manipulate behaviour, or undermine human autonomy are no longer hypothetical risks – they are already emerging in practice. Protecting human dignity and agency must therefore become an immediate governance priority.
  • International coordination should converge around enforceable AI red lines. Far from creating new obligations, red lines apply existing international human rights law to AI. States and companies are already required to respect rights like privacy and equality, and to ensure that any restrictions on rights are lawful, necessary, and proportionate. There is already broad agreement across international bodies and civil society around several areas where red lines are needed, including biometric mass surveillance, social scoring, predictive policing, and AI systems built to manipulate or exploit. What these share is a common outcome: they undermine autonomy and reduce individuals to objects of control.

GPD calls for enforceable red lines to become the minimum common denominator for accountability. We urge states to treat human dignity as an immediate priority, invest in monitoring capacity, and embed limits into global coordination mechanisms and company design practices. 

We call on:

  • Governments: to commit to AI red lines as part of their human rights obligations, help define their scope, and invest in monitoring capacity to enforce them.
  • Dialogue co-chairs and international organisations: to reflect AI red lines in the official outcomes of governance processes, including the Dialogue, and carry them forward through intersessional work.
  • AI International Scientific Panel: to prioritise evidence on high-risk AI capabilities and uses to inform future discussions on AI red lines. 
  • Technology companies: to embed emerging AI red lines into product design by building safety constraints into systems from the outset.

—

Following the Dialogue, GPD will publish a policy brief exploring how red lines can be defined, implemented, and enforced under existing human rights law.