Fide AI

Why now

The case for independent scrutiny.

Independent AI evaluation is becoming an operating practice. Meaningful access, rigorous methods, and independent judgment let outside researchers examine the evidence behind consequential decisions.

Updated

From commitments to working arrangements.

Anthropic ·

Implementation announcement

Anthropic reports employee-like access for METR.

Anthropic reports that six METR staff now have employee-like access to its internal systems. METR can publish assessments without Anthropic’s veto, subject to defined redactions. This makes the earlier commitment concrete, while other evaluator relationships begin with narrower access and publication rights. The arrangements differ by tier; deeper access is not automatic.

Read the source

METR ·

Assessment proposal

METR proposes public assessments of loss-of-control risks.

METR proposes investigations of internal agent alignment, monitoring and control, and acceleration of AI research. Access to staff and records would support public reports explaining findings and limitations. The proposal gives outside scrutiny a concrete set of questions and methods, while explicitly stopping short of confident assurance that development is safe.

Read the source

The demand extends across frontier labs. OpenAI has outlined priorities for independent assessment of safeguards, capability evaluations, safety cases, and critical incidents.

Read its assessment priorities

Public support

A broad call for independent evaluation.

Leaders of AI labs, independent researchers, and others across the field have called for stronger outside scrutiny. Explore their statements, commitments, and invitations to build.

These statements describe their authors’ positions; access and publication terms depend on the actual arrangement.

The call for independent evaluation

01 / 14

Post screenshots saved September 25, 2026.
Public statements, not endorsements of Fide AI.

Fide’s contribution

Turn access into useful evidence.

We welcome this progress. Fide contributes model comparisons, agent workflow assessments, and scrutiny of evaluations themselves. A defined technical contribution can help a lab or evaluation partner investigate a consequential decision, with methods, access, and reporting terms agreed in advance.

Source note: We distinguish reported implementation, proposals, and research findings. Dates refer to the sources; the interpretation is Fide’s. These developments do not imply an endorsement of Fide or an evaluator appointment.