Human Rights and the Regulation of AI — September 2026 → UK AI Security Institute
Recommends compulsory model access
2026-09-14
AISI would review powerful models and receive specifications under statute.
Public research and evaluation
UK AI Security Institute
Government AI research institute and METR partner.
September 2026 update
Government AI research institute and METR partner.
A proposal to replace voluntary model access with a duty
The JCHR report recommends giving AISI statutory access to powerful AI models and their specifications. The report records the minister’s view that voluntary access was sufficient, so the proposed duty is a change the committee seeks, not a power already demonstrated. It also allows AISI and broader oversight functions to be combined.
What the records show
Joe Benton → UK AI Security Institute
Helped establish predecessor taskforce
Checked 14 September 2026; event date unresolved
Reports helping set up the UK Frontier AI Taskforce.
UK AI Security Institute → METR
Acknowledged supporter and partner
Checked 14 September 2026; event date unresolved
METR lists AISI support and partnership.
Further reading
Who evaluates the evaluators? Money, access and the METR network ↗
Read the original sources 2
See the connections
Open this part of the network ↗Rules & government
METR: funding, access and policy
Follow grants, lab access, people and public-sector work around the evaluator.
Open diagram →Rules & government
UK AI oversight: from recommendation to proposed powers
Follow the committee report to two proposed sets of decision rights.
Open diagram →What the connections say
3 relationshipsJoe Benton → Helped establish predecessor taskforce → UK AI Security Institute
Checked 14 September 2026; event date unresolved
Reports helping set up the UK Frontier AI Taskforce.
Read the original source 1
UK AI Security Institute → Acknowledged supporter and partner → METR
Checked 14 September 2026; event date unresolved
METR lists AISI support and partnership. Amount and terms unspecified.
Read the original source 1
Human Rights and the Regulation of AI — September 2026 → Recommends compulsory model access → UK AI Security Institute
2026-09-14
AISI would review powerful models and receive specifications under statute. Recommendations 231–233; AISI would advise the oversight decision-maker. Functions might be combined. This is proposed, not current compulsory authority.