In autumn 2025, the UK government concluded a Spending Review that allocated £240 million to artificial intelligence safety and security initiatives—a significant expansion of state capacity in a sector where voluntary codes have dominated policy for three years. Equally important: the body formerly known as the AI Safety Institute (AISI) was officially renamed the AI Security Institute (AISI), signalling a strategic pivot toward national security framings alongside safety research.

For Chief AI Officers and senior technology leaders in the UK, this transition marks a critical inflection point. The shift from "safety" to "security" language, combined with substantial funding, suggests the government is moving toward more prescriptive governance pathways. Voluntary standards—the cornerstone of the UK's 2023–2024 AI regulation stance—now sit alongside growing state institutional capacity to assess, benchmark, and potentially enforce AI risk controls.

This article unpacks what has changed, why it matters for UK enterprises, and how CAIOs should interpret the governance landscape as 2026 unfolds.

The Rename: From Safety to Security

The rebranding from AI Safety Institute to AI Security Institute is not semantic repositioning alone. It reflects a maturing understanding within UK policy circles that AI risks extend beyond technical robustness and bias mitigation into national security, critical infrastructure resilience, and information warfare vectors.

Under the previous safety framing, the institute's remit centred on frontier model testing, alignment research, and support for industry self-regulation through standards bodies like the British Standards Institution (BSI). The security framing broadens mandate scope to include:

• Adversarial resilience: How AI systems withstand deliberate attacks, data poisoning, or manipulation by hostile actors.
• Supply chain integrity: Governance of training data provenance, model weights custody, and compute infrastructure.
• Information security: Prevention of AI-enabled deepfakes, synthetic media, and coordinated inauthentic behaviour at scale.
• Critical national infrastructure: AI deployment in energy, water, transport, and defence systems.

This mirrors language and priorities reflected in the UK government's AI regulation and standards collection, which emphasises "trusted AI" frameworks aligned with national interests.

The £240 Million Spending Review Settlement

The 2025 Spending Review allocated £240 million to AI safety and security work across government, with the AI Security Institute positioned as a central recipient and operational hub. This represents a 4–5x increase on prior annual budgets and signals genuine state commitment to building institutional capacity.

Allocation priorities within that settlement appear to include:

1. Frontier model evaluation: Infrastructure and expertise to conduct red-teaming, adversarial testing, and capability assessment of large language models and multimodal systems developed in the UK and internationally.
2. Standards research and benchmarking: Development of risk assessment frameworks compatible with international standards harmonisation efforts, particularly alignment with EU AI Act risk classifications and emerging OECD recommendations.
3. International coordination: Participation in forums such as the Bletchley Declaration signatories and the AI Safety Institute Consortium (AISIC), ensuring UK research informs global norms.
4. Workforce and expertise: Recruitment and retention of AI researchers, policy analysts, and technical staff to build capability in-house rather than relying solely on external partnerships.

The scale of this investment reflects an implicit government assessment that the voluntary code era has limits and that institutional capacity—the ability to independently evaluate and audit AI systems at scale—is now a national priority.

Voluntary Codes Transitioning to Structured Oversight

The UK's regulatory strategy since 2023 has rested on principles-based guidance and industry-led codes of conduct rather than prescriptive statute. Organisations like the Ada Lovelace Institute and the Alan Turing Institute published frameworks; the BSI developed draft standards; industry consortia drafted voluntary principles.

The 2025 Spending Review settlement and AI Security Institute expansion do not abolish this approach. However, they create structural conditions for transition:

• Benchmarking pressure: As the institute publishes risk assessment data and model evaluation results, voluntary codes will face implicit—then explicit—expectations to align with published standards.
• Regulatory precedent: Early enforcement or compliance actions by regulators (ICO, FCA, sector-specific bodies) will reference institute research, creating de facto binding force.
• International harmonisation: The EU AI Act's mandatory compliance regime (in effect 2025–2026) will force UK businesses to choose between dual compliance or accepting EU legal risk. The institute's work on international standards may influence whether the UK adopts EU risk classifications or develops distinct frameworks.
• Procurement leverage: Government contracts increasingly require AI risk assessments aligned with institute guidance; this cascades through supply chains.

For CAIOs, this means the comfortable assumption that "we comply with the voluntary code" may not insulate organisations from future regulatory exposure.

Partnerships, Capacity, and Institutional Positioning

The scope and organisational location of the AI Security Institute within UK government remains a live question. It operates under the Department for Science, Innovation and Technology (DSIT) but has unclear formal authority to mandate compliance or conduct inspections. This gap is deliberate: the institute is positioned as a research and advisory body, not a regulator.

Partnerships with academic institutions (particularly the Alan Turing Institute), defence and security bodies (GCHQ, MOD AI), and international counterparts (e.g., the US AI Safety Institute and similar initiatives in Canada, France, Germany) provide research capacity and legitimacy without duplicating regulatory powers.

The risk for UK enterprises: partnerships and research collaborations can inform regulatory change without formal notice. CAIOs should monitor publications and policy briefs from the Alan Turing Institute and DSIT-commissioned research as leading indicators of likely governance shifts.

Standards Alignment and the EU AI Act Spillover

A critical pressure point for UK AI governance is the EU AI Act. Although the UK left the EU, the Act's extraterritorial scope means any UK business operating in the EU, selling to EU customers, or using EU data must comply. This creates regulatory arbitrage pressure: why develop dual compliance systems?

The AI Security Institute's work on international standards aims (in principle) to harmonise UK and EU approaches without the UK adopting the Act wholesale. Government guidance on AI assurance techniques uses risk-based language compatible with the Act's framework, suggesting convergence even in absence of formal legal harmonisation.

For CAIOs:

• Assume UK companies will face de facto EU-aligned risk classification systems regardless of formal statutory divergence.
• The institute's published risk assessment methodologies will likely become the reference standard for UK regulators.
• Early adoption of institute-recommended evaluation techniques creates competitive and compliance advantage.

What CAIOs Should Do Now

The governance landscape is shifting from voluntary codes toward structured, evidence-based oversight. Practical steps:

1. Monitor institute publications: Subscribe to DSIT communications and the AI Security Institute research outputs. These documents signal regulatory direction months or years before statutory change.
2. Audit AI risk assessment practices: If your organisation's risk assessment process does not align with institute guidance on frontier model testing, bias evaluation, and adversarial resilience, begin alignment work now.
3. Engage with standards bodies: The BSI and other standards-setting bodies are now aligned with government priorities. Participation in standards development (through committees or public consultation responses) shapes outcomes.
4. Plan for international compliance: If operating in the EU, assume future UK regulation will converge toward EU standards. Dual compliance is expensive; early harmonisation is cheaper.
5. Build internal expertise: The government is recruiting AI security specialists. Competition for talent will increase. Invest in capability-building and retention now.

Forward-Looking Analysis: What 2026–2027 May Bring

Based on current trends, several plausible scenarios emerge:

Scenario 1: Structured Voluntary Codes Become Binding
Government endorses a formal industry standard (likely BSI or equivalent) and conditions public procurement, export licensing, and regulatory forbearance on compliance. Functionally binding without statutory force.

Scenario 2: Sector-Specific AI Regulation
Rather than comprehensive AI law, the UK adopts targeted regulation in high-risk sectors: financial services (FCA), healthcare (CQC/NHSX), critical infrastructure (Ofgem, Network Rail). The AI Security Institute provides evidence base.

Scenario 3: Positive AI Regulatory Framework
The government legislates a high-level AI Act modelled loosely on the EU approach but with lighter compliance burden. The institute's risk assessment work becomes a statutory reference standard.

Scenario 4: International Standard-Setting Leadership
The institute's research informs global standards (ISO/IEC, NIST, ITU), making the UK a standard-setter rather than a rule-taker. UK businesses benefit from early alignment with eventual global norms.

All scenarios converge on one point: organisations that have aligned their AI governance with institute recommendations and evidence-based risk assessment will face lower friction, lower cost, and lower reputational risk as regulation evolves.

Conclusion: From Principles to Evidence

The UK's AI governance model is maturing from principles-based self-regulation toward evidence-driven, standards-aligned oversight. The AI Security Institute's expansion and rebranding reflect this shift. The £240 million Spending Review settlement funds institutional capacity to assess and benchmark AI systems at national scale—the precondition for effective regulation.

For CAIOs, this is not a crisis; it is a signal to act. Organisations that invest now in robust risk assessment, international standards alignment, and evidence-based AI governance will be positioned to shape rather than react to regulatory change. Those that wait risk costly retrofits and regulatory friction.

The voluntary code era is not over, but its days are numbered. The governance landscape of 2027–2028 will be defined by what the AI Security Institute learns and publishes in 2026. Pay attention.