UK AISI vs US CAISI: AI Safety Institute Compared (July 2026)
UK AISI vs US CAISI: AI Safety Institute Compared (July 2026)
The July 20-21, 2026 window crystallized a divergence that had been building for 18 months: UK AI safety governance is now materially stronger than US AI safety governance, and the gap is widening.
On July 20, 2026, US CAISI director Chris Fall resigned after three months. On July 21, 2026, UK Prime Minister Andy Burnham appointed Kanishka Narayan as the UK’s first cabinet-level AI Minister. Together, these events mark a structural shift: the UK now holds Western leadership on frontier AI safety governance.
Last verified: July 21, 2026
Head-to-Head Table
| Dimension | UK AISI | US CAISI |
|---|---|---|
| Launched | November 2023 (Bletchley Park summit) | 2024 (as USAISI, rebranded to CAISI in 2025) |
| Parent agency | Department for Science, Innovation and Technology (DSIT) | NIST → Department of Commerce |
| Staff (July 2026) | ~90+ | ~30-50 |
| Leadership stability | Ian Hogarth (stable, ~2.5 years) | 3 leaders in 12 months (Kelly, Burns, Fall) |
| Current leader | Ian Hogarth | Arvind Raman (acting, dual-hat with NIST) |
| Cabinet-level advocate | ✓ Kanishka Narayan (from July 21, 2026) | ✗ (OSTP AI advisor only) |
| International MOUs | Japan AISI, EU AI Office (active) | Slowed since 2024 |
| Budget trajectory | Expected to double by 2027 | Uncertain |
| Political weight | High (cabinet-level) | Moderate (buried in Commerce) |
| Frontier lab engagement | Active pre-release evaluations | Slowing due to leadership churn |
Origin Stories
UK AISI: Born from the Bletchley Park Moment
November 2023: Rishi Sunak’s Conservative government hosts the UK AI Safety Summit at Bletchley Park. 28 countries + EU sign the Bletchley Declaration on frontier AI risks. UK announces the AI Safety Institute the same week.
Ian Hogarth appointed — angel investor and AI observer with deep tech industry ties. Given operational independence within DSIT.
Early wins (2024):
- Voluntary pre-release evaluation MOUs with OpenAI, Anthropic, DeepMind.
- Publication of INSPECT evaluation framework (open source).
- International coordination with Seoul AI Safety Summit (May 2024), Paris AI Action Summit (Feb 2025).
Continuity through Labour transition (2024): Keir Starmer’s July 2024 Labour victory preserved AISI’s independence and expanded funding. Ian Hogarth continued as director — political continuity across a change of government is unusual and consequential.
US CAISI: Institutional Instability from the Start
2024: US AI Safety Institute (USAISI) established under Biden administration at NIST. Elizabeth Kelly appointed director.
Q1 2025: Trump administration takes office. Rebrands USAISI to CAISI (Center for AI Standards and Innovation) — emphasis shifts from “safety” to “innovation” in framing.
Q2 2025: Elizabeth Kelly departs; Collin Burns appointed director. Burns removed shortly after appointment due to prior Anthropic work.
April 2026: Chris Fall appointed director.
July 20, 2026: Chris Fall resigns after three months. Arvind Raman (NIST director) named acting director.
Cumulative result: three directors in 12 months, ongoing budget uncertainty, no cabinet-level AI Minister equivalent, weakened political position.
Head-to-Head on Key Dimensions
Staff Size
| Institute | Staff (July 2026) | Trajectory |
|---|---|---|
| UK AISI | ~90+ | Growing to ~180 by 2027 |
| US CAISI | ~30-50 | Uncertain |
Winner: UK. 2-3x larger today, growing faster.
Leadership Stability
| Institute | Leaders since 2024 | Current status |
|---|---|---|
| UK AISI | 1 (Ian Hogarth) | Stable |
| US CAISI | 3 (Kelly, Burns, Fall) | Acting director dual-hatted with NIST |
Winner: UK by a lot. Leadership stability is prerequisite for institutional credibility.
Political Weight
| Institute | Cabinet-level advocate? |
|---|---|
| UK AISI | ✓ Kanishka Narayan (AI Minister, July 21, 2026) |
| US CAISI | ✗ (no cabinet-level AI role) |
Winner: UK. Cabinet-level advocacy protects budgets and independence.
Frontier Lab Engagement
| Institute | Active pre-release evaluations |
|---|---|
| UK AISI | OpenAI, Anthropic, DeepMind, xAI (ongoing) |
| US CAISI | Slowing due to leadership churn |
Winner: UK. Labs submit models when institute is functional; leadership churn reduces submission cadence.
International Coordination
| Institute | Active MOUs |
|---|---|
| UK AISI | Japan AISI, EU AI Office, participates in G7 AI process |
| US CAISI | International engagement slowed since 2024 |
Winner: UK. International network multiplies institutional influence.
Publication Cadence
| Institute | Published evaluations 2026 |
|---|---|
| UK AISI | ~15-20 evaluation reports + methodology papers |
| US CAISI | ~5-8 evaluation reports |
Winner: UK. Publication cadence signals institutional productivity.
Why the Divergence Happened
1. Political Prioritization
UK: Both Conservative (Sunak, 2022-2024) and Labour (Starmer, 2024-2026; Burnham, 2026-) governments have treated AI safety as strategic priority. The Bletchley Declaration was a signature Sunak initiative; Starmer preserved it; Burnham elevated it to cabinet rank.
US: Trump administration deprioritized AI safety in favor of AI competitiveness. Removal of Biden-era AI Executive Order provisions signaled the shift. CAISI continues to exist but with less political weight.
2. Leadership Continuity
Ian Hogarth’s stable leadership at UK AISI allowed institutional culture, evaluation methodology, and international relationships to develop consistently over ~2.5 years.
US CAISI’s three leaders in 12 months repeatedly reset institutional priorities, disrupted staff, and interrupted frontier lab engagement.
3. Institutional Design
UK AISI has more independence within DSIT — direct reporting to the AI Minister, budget line item, clear mandate.
US CAISI is buried inside NIST inside Commerce — three layers of institutional hierarchy before you reach the Secretary of Commerce. Political weight is diffused.
4. Cultural Factors
UK tech policy culture has historically valued expert-led institutions (Ofcom, CMA, ICO). AISI fits this pattern.
US tech policy culture has historically preferred market-driven approaches with lighter regulatory bodies. CAISI doesn’t fit as naturally into US institutional norms.
What This Means for AI Industry
Frontier Labs
Primary implication: UK AISI becomes the reference international pre-release evaluation authority.
- Voluntary submissions to UK AISI become more valuable as a safety credential than submissions to US CAISI.
- UK AISI evaluation methodologies (INSPECT framework, others) become de facto standards.
- Model behavior tuning may prioritize UK AISI test criteria over US CAISI criteria.
Enterprise AI Buyers
Practical implication: UK AISI evaluations become the emerging credibility signal for AI safety claims.
- When evaluating AI vendors, check whether they’ve completed UK AISI evaluations.
- UK AISI publications (safety assessments, red-teaming results) become useful due diligence resources.
- US CAISI evaluations are less reliable near-term due to institutional weakness.
AI Startups
Location-dependent implications:
UK-based startups:
- Engagement with UK AISI is strategically important — establish relationships early.
- UK Global Talent visa (expected expansion under Narayan) eases hiring.
- Government AI procurement opportunities expand.
US-based startups:
- Watch UK AISI evolution to anticipate regulatory direction.
- Consider UK evaluation submissions for international credibility.
- Prepare for potential UK AI Regulation Bill (2026-2027) affecting UK operations.
International AI Governance
Center of gravity shifts modestly from Washington to London. The US retains most of the frontier labs, but the UK increasingly sets the safety norms and evaluation methodology those labs must engage with.
Expected downstream effects:
- G7 AI process places more weight on UK AISI outputs.
- EU AI Act implementation references UK AISI methodologies.
- Japan and Australia AISI variants model on UK institutional design.
- Global South AI governance discussions look to UK as democratic-world alternative to US market-driven or Chinese state-directed models.
Can US CAISI Recover?
Yes, but it requires deliberate reinvestment. Steps that would restore US CAISI parity with UK AISI:
- Stable leadership — appoint a permanent director with expected 3-5 year tenure. Insulate from political churn.
- Cabinet-level advocacy — create a Presidential AI Advisor role with cabinet rank, or reorganize CAISI as independent agency reporting to White House.
- Budget commitment — multi-year funding at ~$150M/year (roughly matching UK trajectory).
- Staff growth — target 150-200 staff by 2027 to match UK.
- International MOU reactivation — restore bilateral engagement with UK AISI, Japan AISI, EU AI Office.
Political viability: low in the current administration. Recovery is more likely in 2028+ under a subsequent administration (if AI safety governance is prioritized).
Alternative path: consolidate AI safety oversight in a new agency (e.g., proposed “US AI Authority”) rather than trying to fix CAISI’s structural weaknesses. Requires congressional action.
Bottom Line
As of July 21, 2026, UK AISI is materially stronger than US CAISI on every measurable dimension — staff, leadership stability, political weight, publication cadence, international coordination.
The July 20-21 window (Chris Fall resignation + Kanishka Narayan cabinet elevation) crystallized a divergence that had been building since Trump took office in January 2025.
For AI industry: engage with UK AISI as the emerging authority. UK evaluations become the credible international benchmark. US CAISI remains relevant but with weakened weight.
For AI safety community: the US federal path is difficult in the near term. Effort may shift toward UK AISI expansion, EU AI Act implementation, and international coordination that increasingly runs through London rather than Washington.
For the broader arc: the UK has made a strategic bet that AI safety governance will be a defining domain of the next decade — and it’s positioning itself to lead the Western democratic world on it. Ian Hogarth’s institutional building and Kanishka Narayan’s cabinet elevation are the concrete signals. Watch whether the US responds structurally, or accepts the shift.
Sources
- UK AI Safety Institute official page: aisi.gov.uk
- UK AISI INSPECT evaluation framework: github.com/UKGovernmentBEIS/inspect_ai
- US CAISI at NIST: nist.gov/aisi
- Business Insider Chris Fall CAISI resignation coverage: businessinsider.com
- Indian Express Kanishka Narayan UK AI Minister profile: indianexpress.com