Intelligence Made Human: What Global AI Governance Means for Regulated Industries

At Marino, we’ve long believed that technology only earns trust when it’s built with people at the centre. “Software Made Human” captured that. But software was always the medium. It is the application layer.

What regulated industries are now grappling with directly, is intelligence. How it reasons, makes decisions, and acts in the world. That’s the core layer we’re working in. AI is the material we are now working with, and “Intelligence Made Human” is our focus.

Right now, nowhere is that idea more visible than in AI governance. As we work with clients across financial services, healthcare, and the public sector to deploy AI responsibly, understanding the global regulatory landscape is core to how we think and operate.

It’s also central to Evident, a platform for compliant and governed AI deployment in regulated industries, created by the Marino team.

Here’s how the major frameworks compare and what they mean in practice.

1. EU AI Act: the gold standard

We previously posted about how the EU AI Act is the world’s first comprehensive AI legislation. Enforcement began April 2, 2025.

For agentic AI, the key takeaways:

• Risk-based approach with four tiers
• Strict requirements for high-risk AI (compliance, logging, human oversight)
• Bans on unacceptable-risk systems (e.g. social scoring, subliminal manipulation)
• GPAI model transparency obligations
• Significant penalties (up to 7% global turnover)

2. US: state-by-state and NIST frameworks

The US takes a more fragmented approach.

Federal level

• White House Executive Order on AI (Oct 2023): requires safety testing for foundation models, encourages voluntary standards, prioritises cybersecurity.
• NIST AI Risk Management Framework: provides guidance but is not legally binding. The 2.0 update adds agentic-specific guidance on autonomous-action risk.
• AI Bill of Rights: non-binding principles around algorithmic discrimination, privacy, notice and explanation.

State level

• California: SB 1047 (Safety and Accountability for Frontier AI Models), not currently in law, a risk-based approach (paused following legal challenges).
• Colorado: AI consumer protection bill requiring disclosure of AI use in consumer interactions.
• Oregon: algorithmic transparency requirements for high-stakes decision-making.

For agentic AI, the lack of federal uniformity creates complexity. If you’re building agents for US deployment, you need to navigate both sectoral regulations (e.g. healthcare HIPAA, financial regulations) and emerging state laws.

3. China: generative AI measures

China has taken one of the more prescriptive regulatory approaches:

• Interim Measures for Generative AI (2023): requires training data alignment with socialist core values, content filtering, and real-name registration for users.
• Deep Synthesis Provisions (2023): focus on labelling AI-generated content and preventing misuse.
• Algorithmic Recommendation Measures (2022, updated 2024): targets recommendation algorithms, requires transparency, and mandates user opt-out options.

The data sovereignty requirements are particularly relevant for agentic AI. Any agent operating in China must comply with data localisation requirements, government alignment and content review obligations, and security assessment requirements for models with public influence.

4. UK: “pro-innovation” approach

The UK’s self-styled pro-innovation, non-statutory approach centres on five principles, safety, transparency, fairness, accountability, and contestability, with sector-specific regulators (CMA, ICO, MHRA) overseeing AI in their respective domains. The AI Safety Institute leads world-class frontier AI testing, particularly relevant for agentic systems.

For agentic AI, financial services agents are regulated by the FCA, healthcare agents by the MHRA, and so on. It is light-touch at the framework level. But sector-specific regulation can still carry real weight.

5. Singapore: Model AI Act

Singapore has been proactive with its Model AI Act (2024), the first draft framework of its kind. It takes a risk-based approach similar to the EU, but is more guidance-oriented, with practical tools like AI Verify for testing AI systems. The focus on transparency, fairness, and accountability makes it one of the more developer-friendly frameworks globally.

6. Brazil: AI Bill of Rights

Brazil’s AI Bill of Rights (PL 2338/2023) is notably human-centred, emphasising transparency, non-discrimination, and the right to human review. A new national AI regulatory authority (SANAI) will have enforcement powers. For agentic AI, the emphasis on human review rights and transparency aligns closely with how we approach responsible deployment at Marino: the human always needs to remain meaningfully in the loop.

Comparison of frameworks

Jurisdiction	Approach	Enforcement strictness	Innovation support	Agentic AI impact
EU	Comprehensive legislation	Very high	Medium	Strict compliance requirements
US	Fragmented, state + NIST	Medium	Very high	Complex multi-jurisdictional navigation
China	Prescriptive, data-driven	Very high	Low	Data sovereignty requirements
UK	Principles + sectoral	Medium	Very high	Sector-specific oversight
Singapore	Risk-based, guidance	Low–medium	Very high	Practical compliance tools
Brazil	Bill of Rights	Medium–high	Medium	Human-centred design focus

7. Multi-jurisdiction checklist

If you’re deploying agentic AI globally, your checklist expands:

• Map jurisdictions. Identify where you operate and which laws apply.
• Classify per region. Risk classification may differ by jurisdiction. A system could be High Risk in the EU but not in the UK.
• Data compliance. Data location, transfer, and sovereignty requirements vary significantly (e.g. GDPR in EU, PIPL in China, LGPD in Brazil).
• Content requirements. Content standards and moderation vary by region.
• Transparency obligations. Disclosure requirements vary.
• Establish local counsel. Legal expertise is essential for multi-region deployments.
• Design for the highest standard. If feasible, design your agentic systems to meet the most stringent standard (currently the EU AI Act) to simplify compliance.
• Monitor regulatory developments. The regulatory landscape is evolving rapidly.
• Build governance into architecture. Governance should be built into your agentic systems, not bolted on afterwards.
• Train teams. Teams need to understand region-specific obligations.

Global agentic AI deployment requires multi-jurisdictional competence. The EU AI Act sets a high bar. Many other countries are developing their own frameworks.

At Marino Software, we are building platforms specifically to help regulated-sector organisations navigate this complexity, so that compliance becomes an enabler of confident AI adoption. Building in governance from the start is our default. It avoids costly remediation later. Intelligence Made Human means building that trust from the ground up.

Sources and references

EU AI Act · NIST AI Risk Management Framework 2.0 · China Interim Measures for Generative AI · UK AI Regulation: A Pro-Innovation Approach · Singapore Model AI Governance Framework · Brazil AI Bill