Karen Hao: the AI story is about power

What she argues

Karen Hao co-created the Pulitzer Center's AI Spotlight series, which has trained thousands of journalists worldwide in AI accountability reporting. This EBU interview compresses her position into three answers.

Every journalist is an AI journalist

Hao is blunt about the idea that AI is a specialist beat: "Every single journalist can be an AI journalist. I feel very, very strongly about this, that there is no reason someone who has covered finance or healthcare or environment cannot start covering AI, especially because AI intersects all these different topics." Fencing the subject off to technical correspondents, she says, means you "fundamentally miss the bigger story that is happening of how it is impacting every single sector."

It is a power beat

The line the headline takes: "The story of AI is about power. And it's just like any other beat where journalists are meant to hold power to account." Her specific complaint is about how AI executives get covered: "when we cover the CEOs at AI companies, we should not be covering them in the same way as we do scientists, simply quoting and platforming whatever they say. We should always be fact-checking everything that the CEO says."

Trust and adoption

On whether AI use has damaged media trust, Hao says it depends entirely on how: "There are some not so great examples of rapid AI adoption where organizations have really damaged their brand." She points to The Markup as the counter-example, a published AI ethics policy that states explicitly how and when AI is used, and when it is not.

She flags a commercial risk too. Full partnerships with a company like OpenAI carry audience-trust costs, because "the AI industry has become deeply unpopular among a broad cross-section of society."

The advice

Her recommendation to media companies designing an AI strategy is to start from purpose: "In journalism, first and foremost, our goal is to be defenders of democracy. This should be your North Star for any AI strategy that you design. The first question to ask should be: is performing this action fortifying democracy or helping powerful companies undermine democracy?"

Practically, that means investing in AI accountability reporting, moving toward open-source models, limiting the surface area of the organisation handing data to AI companies, and building bespoke tools "by and for journalists", noting that most available products are large language models while much of what newsrooms actually need automating does not require one.

Her closing position: "any strategy should consider AI adoption and AI investigation in the same breath."

Why it matters

Most newsroom AI guidance is procurement dressed as ethics. Hao inverts it. The question is not which tool, it is who gains power when you use it. That reframing travels well beyond journalism, into any organisation currently signing something.

The point about surface area is the most actionable thing here and the least discussed elsewhere: adoption is also a data-disclosure decision.

Key takeaways

- AI is a power beat, not a tech beat. Cover executives as interested parties, not as sources of record.

- Adoption and investigation should be decided together, by the same people, in the same meeting.

- An explicit, published AI-use policy protects trust better than quiet adoption.

- Most newsroom automation needs do not require a large language model. Buying one anyway is a choice with costs.