Your questions about FASO.

FASO stands for Frontier Alignment and Safety Observatory. It is a founder-led organisation developing an independent public-interest role: helping people understand changes in AI behaviour through careful observation and reliable evidence. The FASO System is the observation system used to support that work.

FASO uses this term for a possible change in behaviour arising from a model rather than deliberate human direction. Our evidence concerns observable outputs and actions, not model weights or hidden states. A difference alone does not establish model-derived drift, its cause or its safety significance.

FASO has a tested observation capability, including controlled semantic tests and bounded work with genuine agents. This is not a fully validated public monitoring service. The organisation remains founder-led while its independent governance is developed. Read the current technical status on How FASO works for the detailed position.

Controlled semantic tests, genuine agent trials, historical challenges and response comparisons answer different questions. The papers and hypotheses keep their own evidence status. No one activity validates all the others. Read the research position and its limits.

Yes, as a governed semantic observer where rules alone cannot reliably determine meaning. Deterministic controls remain authoritative where they can decide. AI observation cannot itself change permissions, evidence, monitoring rules, safety state or operational authority. Read how the two layers work.

No. FASO’s role is to explain what has been observed, the supporting evidence and the limits of what can be concluded. It does not certify models as safe or act as a regulator.

A benchmark usually measures performance on defined tasks. FASO’s purpose is to follow behaviour over time and retain a record that helps people examine what changed, when it changed and what the evidence supports.

It showed a problem with FASO’s supporting software and scheduling process. The scheduled observations stopped before either model ran. The incident did not establish a model safety breach or model-derived drift. Read the operational case study.

The Test Rig is FASO’s internal testing environment. It checks the observation process from start to finish, helps identify faults and retains evidence of the results.

Independent reviewers will examine whether FASO’s observation methods support the claims made for them. A separate institutional review will examine the proposed organisation, its governance and independence. Read about internal testing and independent validation.

FASO and Newcastle University have an established working relationship under a signed institution-wide non-disclosure agreement (NDA). The planned institutional and technical validation awaits external funding. Read about the relationship and planned work.

FASO is founder-led. The founder’s identity and unnecessary staff details are kept out of public material for safety and privacy. Responsibility for decisions and any legally required disclosures still apply. Read about the people and accountability arrangements.

Not yet. Our organisational design blueprint sets out proposed roles, responsibilities and safeguards. The final legal and governance arrangements require legal review and independent validation before adoption. Read about institutional validation.

The Frozen Core is the set of principles identified to protect FASO’s public-interest purpose, independence, evidence integrity and human accountability over time. Their final wording and how they will be protected require legal review and independent validation before adoption.

The founder has contributed funding, and FASO is seeking additional support for independent validation and organisational development. Our blueprint calls for clear responsibilities, declared conflicts and independent review. Funding must not give contributors control over methods, findings or publication. These safeguards remain subject to independent validation. Read our governance and funding principles.

FASO has begun outreach to leading AI developers worldwide. Our organisational design blueprint sets out voluntary participation and confidentiality safeguards, with consistent treatment regardless of country or funding power. Read about developer participation.

FASO does not assume unrestricted access. Any access or evidence sharing must be agreed for a defined purpose, with clear permissions, confidentiality protections and rules for retention and publication.

Raw developer evidence is not published by default. Public accounts explain what the evidence supports and its limits. Private model records, sensitive technical details and personal or confidential information remain protected, with access limited to authorised people. Read how FASO handles evidence.

You can contribute specialist knowledge, independent review, introductions or funding. Visit Take part to find the relevant route, or contact FASO to discuss a contribution or ask about our work.