As the rise of AI adoption and development continues, so does its risk assessment and regulatory recommendations. It’s not lost on professionals that while the promises of AI’s potential have been abundant, so have the warnings of using it without proper guardrails.
There are regulations and frameworks that assess the risks and make informed recommendations: as the first intergovernmental standard on AI, the OECD AI Principles “promote innovative, trustworthy AI that respects human rights and democratic values.” The OECD AI Principles were last updated in 2024, with OECD member countries and numerous global partners adopting them, and recommendations of practical guidance on AI.
The EU AI Act assigns applications of AI to four categories: unacceptable risk; high-risk applications; and applications considered limited, or minimal risk, which are left mostly unregulated.
Broadcast and live production technology mostly falls into the last two categories, but the multiple ways AI can play a part live in production workflows is only growing and we must be committed to navigating that growth securely.
AI adoption: dialling up and down
Being practical, there’s no one-size-fits-all application for AI in the production process. For a tool to be useful, it has to be flexible, giving broadcasters the option to dial it up and down to best fit the team’s way of working. That’s the difference of having a human across the loop; the ability to have oversight and control at every point of the production as you choose.
Customising AI adoption can take many shapes. For instance, let’s say a production team wants to automate as much of their show as possible – so, pulling in the story, selecting imagery, perhaps even drafting the piece for broadcast. They can train AI agents to understand their brand and tone of voice, and the type of stories the show tells.
A fully automated workflow can look like this: a journalist writes a few prompts, and then the AI workflow finds the assets that could be associated with the story, writes the story in the journalist’s tone of voice, and puts it straight to air.
A different team might not want that; instead, automating to a point where a human checks it, proofreading the copy and images. Another team might want to automate the copy but do a final check on the assets before a final push.
Those decisions still put the editorial control at the hands of humans. Regardless of the extent each team adopts AI in its workflow, people still own the oversight.
Agentic AI in compliance: self-flagging workflows
With Agentic AI, transparency is still a necessary part of compliance, regardless of the levels of risk.
Categorically, minimal risk includes spam filters or content recommendation engines. We’re used to this, for instance when a streaming platform communicates to its viewers that their data might be used to make recommendations when they’re watching content. Limited risk can include chatbots or AI-generated content, and people need to know when they’re interacting with AI. At this point, we generally know—you see these things every day.
With high-risk activities, there needs to be human oversight, but also audit trails, where the agents themselves can justify the decision-making and show the reasoning. For example, when an agent pulls an image or starts writing copy, it can simultaneously provide the reasons as to why a particular image is appropriate for the story. People can decide where their input is needed for the agent to progress to the next step.
From a regulatory perspective, there are specific controls and transparency duties depending on how the AI being used, whether it’s the agents or the services being provided, is actually rated.
At Vizrt, we’re being very deliberate in our approach with a central hub. This works as an operations centre where the user sets the guardrails – if the next step isn’t within the scope set for the agent, it flags itself, requesting human approval.
How we can promote innovative, trustworthy AI
A significant caution broadcasters have in adopting AI tools is prioritising the trust they’ve built with their audience. But adopting AI into workflows doesn’t require obscurity – across AI regulations and frameworks, the main recommendations are safety and transparency. As AI adoption grows, and as broadcasters remain steadfast in their commitment to transparency, people will become more comfortable with its use.
But it’s a continuous task: authorities should update the regulation as AI adoption progresses and people should remain discerning when interacting with the technology.
