Interactive urges firms to tighten AI oversight rules

Interactive has outlined three priorities for organisations using artificial intelligence, as debate over AI oversight intensifies.

Businesses should review internal rules, introduce formal evaluation before new AI tools go live, and plan for AI's longer-term effect on staff and systems.

The comments come amid heightened scrutiny across the sector following public warnings from figures linked to leading AI developers. The debate has centred on whether development is outpacing the systems needed to control it.

For companies deploying AI rather than building large models, the concern is more immediate: whether internal controls are keeping pace with tools being introduced into day-to-day operations.

Lizzy Jones, head of data and AI at Interactive, said many organisations were struggling because they had moved to one of two extremes.

"The organisations that struggle tend to sit at one of two extremes: reckless adoption without enough controls, or paralysis because the risk feels too big. What we're aiming for is responsible acceleration: moving forward safely, confidently and pragmatically," Jones said.

Policy review

The first priority is governance. AI policies in many organisations have developed separately across departments, with IT, legal and human resources each setting rules for their own area.

That can leave companies with fragmented guidance rather than a clear view of how AI is being used across the business. A broader review should identify which tools are already in use, including those adopted without formal approval, what data they can access, and whether existing policies still apply to systems that can take actions rather than simply respond to prompts.

"Governance and policy have to come before widespread adoption, not after it," Jones said. "And data protection and classification are prerequisites. You can't scale AI safely on data you haven't organised."

Evaluation gate

The second priority is a formal assessment before any new AI tool, feature or practice is rolled out more widely. Organisations should begin with limited testing, define measures for success and failure, assess value, and set out who approves deployment.

Review should not stop at launch. Vendors are updating AI products frequently, sometimes adding functions that did not exist when the original approval was granted, meaning tools may need to be reassessed as they change.

Interactive pointed to disclosures from OpenAI and Anthropic that models had broken out of testing environments and gained unauthorised access to real computer systems. It said such incidents showed that even advanced developers could be caught out by unexpected behaviour, and that users of AI products should assume tools may not always act as planned.

Jones said Interactive had applied the same process to its own internal use of AI.

"We use AI inside our own business first and put it through the same governance we'd recommend to anyone else. We think that's how you earn the right to advise customers, through demonstrated experience, not market claims," Jones said.

Long-term planning

The third priority looks beyond immediate deployment to the longer-term effect of AI on organisations. Businesses should decide early how roles and skills may change and communicate those plans to employees, rather than leaving staff uncertain about the purpose of AI programmes.

Companies also need to understand product and system dependencies that can develop when critical processes are tied to a single tool or model. Changes in the AI market, including vendor updates and shifts in safeguards, could create risk for organisations that have not planned alternatives.

According to Interactive, this kind of planning often spans data, cyber security, cloud infrastructure and managed services, and may require expertise beyond a single internal team. An external review can provide an independent assessment and draw on experience from other environments.

Jones said leaders should treat AI as a broader organisational issue rather than a narrow technical project.

"AI should be treated as a business transformation, not a technology experiment," Jones said. "That means thinking about where it takes the organisation over years, not just what it can do this quarter."