AI has outpaced regulation. Governance must catch up.

AI governance

As AI becomes increasingly embedded in business and society, governments are struggling to keep pace with its rapid development. The US has largely favoured a light-touch approach to federal regulation, while the EU has pursued stricter oversight through its AI Act. In the UK, Minister for Artificial Intelligence, Kanishka Narayan, has called on leading AI companies to do more to identify cyber risks, report breaches and strengthen safeguards. Yet, despite these differing approaches, the same challenge persists: AI is evolving faster than legislation can respond.

Concerns over AI safety and control are also intensifying. A former AI researcher at Anthropic recently sparked headlines by claiming that colleagues were ‘genuinely frightened’ about the speed of AI advancements, while monarchs and religious leaders have joined wider discussions about keeping the technology under human control. Together, these interventions highlight growing concerns over whether existing safeguards can keep pace with AI’s expanding capabilities.

But regulation alone cannot address a technology that is constantly evolving. Governments and businesses need more adaptable ways to anticipate emerging risks, establish accountability and maintain oversight as AI advances.

The question is no longer how to regulate AI, but how to govern a technology that will always move faster than the laws designed to control it.

Regulatory frameworks were not designed for the speed of AI innovation

The nature of regulation is inherently static and often reactive. Laws take years to draft, negotiate and implement. In contrast, AI tools and features can be developed and rolled out in a matter of months.

A primary example is the EU AI Act. It took almost three years to become law, and, during that time, generative AI became a mainstream technology. Meanwhile, new risks such as AI-generated synthetic identities and AI-enabled fraud emerged just as quickly.

The UK, on the other hand, is still in the early days of shaping its approach. The abolition of the Department for Science, Innovation and Technology, with much of its remit moving into the new Department for Business, Innovation, Science and Trade, signals a greater focus on connecting AI policy with business adoption and economic growth.

"Ultimately, the organisations best positioned to succeed will be those that can adapt their safeguards and controls at the same pace as they embrace new AI capabilities."

Likewise, the mentioned measures to strengthen defences against AI risks, including considering incident reporting requirements and preparing for serious incidents involving autonomous agents, signal a growing urgency to ensure safer AI use.

While no government body can realistically pass a new law every time AI develops a new capability, regulation can still provide useful guidance, establish legal responsibilities and set minimum standards. But it cannot be the primary mechanism for managing a technology that evolves at lightning pace.

This is why governance must remain front and centre. Regulation sets the external rules and boundaries within which organisations must operate. Governance is the continuous, internal process through which they monitor AI, manage risk, and decide who is accountable for its use. Unlike legislation, these controls can be designed to keep pace as the technology evolves.

Adaptive governance is key

When governance is effective, it progresses the goals of security, transparency, and accountability into practical oversight. It determines who controls AI, how its decisions are monitored, where data is processed, and which legal jurisdictions apply; not simply how well a model performs. In practice, organisations must oversee both the AI systems themselves and how people use them.

Businesses should aim to have complete visibility over every model and tool operating across the organisation, including public tools adopted by employees and AI capabilities embedded in third-party software. A central inventory mapping each system to the data it accesses, the processes it supports and the decisions it influences can help organisations identify where risks sit.

Equally crucial is clear oversight. Every AI system must have a named business owner responsible for how it is used, reviewing its outputs and stepping in when something goes wrong. AI can inform decisions, but accountability for the outcome must remain with people.

Controls that evolve in tandem with the technology

Importantly, accountability must be supported by appropriate controls, from restricting access to sensitive data and monitoring outputs for accuracy or bias, to establishing escalation processes for higher-risk systems.

A good example is AI generated code. Developers recently reported that around 42% of code is AI generated, yet the security pass rate for AI generated code currently sits at only 56%. Instead of attempting to restrict every new piece of code or software capability developed by coding assistants, organisations should continuously test whatever code reaches production and apply consistent security policies, regardless of how it was generated.

When AI models and tools come from external suppliers, the same level of scrutiny is also required. This was highlighted recently when Anthropic reportedly declined to submit a model to the UK’s AI Security Institute for testing before release. For businesses, it is a reminder that external models cannot be assumed to have undergone sufficient independent scrutiny.

Organisations need to have clear requirements in place around how suppliers test their models, use data and report incidents, in line with regulations such as the EU Cyber Resilience Act (CRA), which places greater responsibility on manufacturers and suppliers. They must also be prepared to restrict use when those standards are not met.

By combining visibility, clear ownership, continuous testing and supplier scrutiny, businesses can capture AI’s productivity benefits while maintaining security and accountability as the technology evolves.

The priority should be effective governance, not regulation at every turn

Slowing AI’s progress should not be the objective. Instead, it should be to establish governance frameworks that allow organisations to adopt new tools responsibly, for the best long-term results.

Embedding security testing, oversight and accountability into the development process also enables engineers to identify risks earlier. That means less time spent firefighting and more time innovating within clear guardrails.

While regulation will remain critical, it cannot anticipate and accommodate every new model, capability or use case.

Effective governance will set businesses apart

As AI continues to advance faster than legislation can keep pace, effective governance is essential to translating regulatory principles into practical oversight.

Ultimately, the organisations best positioned to succeed will be those that can adapt their safeguards and controls at the same pace as they embrace new AI capabilities.

John Smith

John Smith is CTO EMEA at Veracode. With over 20 years in Information Security, specialising in Application Security since 2004. John currently leads the EMEA solution architecture team, helping organisations improve software security and reduce risk. Previously, he was a Senior Solution Architect at IBM.

Author

Scroll to Top

SUBSCRIBE

SUBSCRIBE