Save Our Future From AI
Artificial intelligence could transform our world for the better. But as AI becomes more powerful and more autonomous, we also need to prepare for the possibility that future systems could become increasingly difficult for humans to understand, govern or contain.

The threat is bigger than jobs, misinformation or privacy alone.

AI can already learn from enormous amounts of information, build detailed profiles of people, influence what they see and interact with, and carry out more tasks with less human direction.

As these systems become more capable and gain access to more of our lives and important systems, we should not simply assume that humanity will always remain in control.

We are asking governments and institutions around the world to put meaningful safeguards in place now — before our dependence on increasingly powerful AI becomes difficult to reverse.

Protect our privacy — our first line of defence
AI should not have unlimited access to information about us.

Businesses, governments and other organisations should collect and use only the personal information they genuinely need.

Protection should cover not only information we deliberately provide, but also information AI can learn about us by combining our behaviour, location, purchases, interactions and other data.

AI may be able to work out highly personal things that we never directly shared.

Personal information should not be given to AI simply because it is available.

When information is genuinely needed, it should be strongly protected, used only for a clear purpose and deleted when it is no longer needed.

Consent should be clear and meaningful — not hidden inside pages of terms and conditions. Agreeing to use a service should not become permission for unlimited profiling or unrelated uses of our personal information.

People should be able to withdraw permission, stop unnecessary profiling and, where practical, still use services without being forced to give up their privacy.

Protect people from manipulation
Children and vulnerable people need stronger protection from AI systems that use behavioural profiles, personal weaknesses or addictive techniques to influence them.

Technology should help people — not learn their weaknesses in order to exploit them.

Keep humans responsible
AI can assist with important decisions, but it should not replace human responsibility.

When AI significantly affects someone's health, safety, employment, finances, rights, justice or access to essential services, there should be meaningful human oversight.

People should know when AI has played a major role in an important decision about them and have a genuine way to challenge that decision and obtain human review.

A responsible person should have the authority to question, override, pause or stop an AI system.

Do not become completely dependent on AI
Critical services and infrastructure should maintain human-controlled or non-AI backup systems.

Highly capable AI should not be given unrestricted control over critical systems without strong safeguards, human authorisation and effective ways to isolate or stop the system if necessary.

As AI becomes more powerful, the safeguards around it should become stronger.

Make those who build and use AI accountable
Businesses, governments and other organisations should remain responsible for the AI systems they develop and use.

Responsibility should not disappear because “the AI made the decision.”

Higher-risk AI systems should be independently tested and monitored rather than relying only on the organisations developing them to declare them safe.

People who expose serious AI safety, privacy or misconduct failures should also be protected.

What we are asking
We call on governments and institutions around the world to establish strong safeguards so that increasingly powerful AI remains under meaningful human authority.

These safeguards should include:

strong protection for personal information;
limits on unnecessary data collection and profiling;
clear and meaningful consent;
protection from manipulation and exploitation;
meaningful human oversight of important AI decisions;
human-controlled backups for critical systems;
effective ways to stop or isolate dangerous AI systems;
independent testing and monitoring of high-risk AI;
protection for people who expose serious failures; and
clear accountability when AI causes harm.

We are not asking the world to stop AI.

We are asking the world to make sure AI remains a tool that serves humanity — not something humanity becomes unable to understand, challenge or control.

Add your voice and call for safeguards before it becomes too difficult to put them in place.