Launching Version 2.0 of the Three Non-Negotiables Policy Framework
We've published Version 2.0 of Achieving the Three Non-Negotiables for Children’s AI Safety.
The original framework was created to show how three essential protections for children could be turned into meaningful policy. Version 2.0 develops this into a more complete strategic proposal, covering the different points at which risks can arise across AI development, deployment and use.
The Three Non-Negotiables
The updated wording of the Three Non-Negotiables is:
- AI systems must never create or facilitate sexualised images of children.
- AI systems must never cultivate, exploit or sustain emotional dependency in children.
- AI systems must never encourage or facilitate children to harm themselves or others.
The wording has been refined to express the intended outcomes more clearly and reduce the scope for avoidable loopholes. The full proposal explains how these red lines should operate in practice, including the evidence providers should be required to produce, the responsibilities that follow when failures occur and the limits of measures such as disclaimers, age labels and internal safety testing.
Looking further upstream
One of the most important developments in Version 2.0 is its more holistic view of children’s AI safety.
Many AI products used by children sit downstream of powerful general-purpose or foundation models. Safeguards added to an individual application cannot always compensate for dangerous capabilities or weaknesses within the model on which it depends.
Open-weight releases can create further difficulties where powerful capabilities are made widely available and become extremely difficult to recall or contain. Effective child protection therefore has to consider how advanced models are developed, evaluated, accessed and released, alongside the design of the products children encounter directly.
Version 2.0 addresses this through five layers of protection, extending from frontier AI development and foundation models through to consumer products, continuing accountability, remedies, age assurance and distribution.
Children’s safety within the AI arms race
The proposal also places children’s AI safety within the context of the international AI arms race.
Competition between companies and countries can create strong incentives to develop and release increasingly capable systems at speed. Safety work may be treated as a delay, while governments can fear that firm regulation will leave them behind. These pressures make it harder for any individual company or country to act with sufficient restraint.
Children will live longest with the consequences of decisions being made now. Protecting them therefore involves responding to immediate harms while also considering the safety, stability, democracy and human autonomy of the society in which they will grow up.
Version 2.0 calls for international cooperation capable of reducing dangerous competitive pressures. It also makes clear that governments should continue implementing achievable protections at product, deployment and access levels without waiting for a complete international agreement. Each layer can advance independently while contributing to a stronger overall system.
Maintaining a firm standard
The updated proposal retains a simple principle - those developing and providing AI systems should carry the burden of demonstrating adequate control.
Where a severe risk remains uncontrolled or has not been properly assessed, the relevant system or capability should be redesigned, restricted, delayed or withheld from release. Deployment should never continue by default simply because a provider has completed its own paperwork, published favourable safety figures or made a voluntary commitment.
This approach is intended to prevent safety-washing and ensure that compliance reflects real protection for children. Regulators need access to credible evidence, independent evaluation and clear powers to intervene.
Read the updated proposal
Version 2.0 is offered as a suggestion for how the Three Non-Negotiables could be achieved. It is intended to support the development of legislation, regulatory codes, international agreements and technical standards while keeping children’s safety and future firmly at the centre.
Read about the Version 2.0 policy framework and download the full proposal.
You can also visit the main Non-Negotiables campaign page to learn more about the campaign and how to support it.
Or download the full policy proposal here: