How artificial intelligence threatens the news industry’s ability to report on existential risks—including those posed by nuclear weapons and AI

The irony is inescapable. Months after the last nuclear arms control agreement between Russia and the United States expired, tech industry leaders are citing Cold War nuclear verification as a model for regulating the safety of artificial intelligence.

Marina Favaro, leader of the Anthropic Institute, and Jack Clark, a co-founder of Anthropic, have called on their industry to build a “brake pedal” because AI is approaching a frontier where humans may lose control as machines develop themselves. “The world has built verification regimes for other complex technologies,” they wrote in a blog post, citing the Intermediate-Range Nuclear Forces Treaty (INF), from which the United States withdrew in 2019, followed by expiry of the New START arms control pact in 2026.

“We’ve done this before. In the height of the Cold War, under highly tense situations between rivalrous countries, they found ways to stabilize aspects of the nuclear arms race,” Clark told CNN.

But as AI satisfies its rapacious appetite for data by reaping the intellectual commons, its spokespeople seem to ignore an important element of the history of nuclear arms control: It was won not only by exhaustive bargaining, but through public action informed by trusted journalism.

Reporting on a subject area as complex and opaque as nuclear weapons—from which the public is currently separated by a yawning gap—demands dedicated, time-consuming journalism. AI threatens the news industry’s ability to support that kind of reportage—not just on nuclear weapons, but also on other existential risks, including artificial intelligence itself.

“Tech giants strip-mine news websites without permission or compensation. They repackage these stolen goods as their own, siphoning off the audiences and revenue that otherwise would go to the news organizations that created this work,” New York Times publisher A.G. Sulzberger told the World Association of News Publishers in Marseilles, France in June. (The Times has sued a number of AI corporations for violations of its intellectual property rights under US copyright law in the training of their products and ongoing use of its work.)

Sulzberger’s comments were delivered shortly after Google announced “the biggest upgrade to our Search box in over 25 years—now completely reimagined with AI,” emphasising answers to questions that are provided directly through AI, rather than through links to news articles and other sources. Initial users say Google AI gives articles on news sites and research studies short shrift.

“How will AI change the news?” Sulzberger asked his Marseilles audience. “How will those changes affect the information ecosystem that serves as the public square for engaged citizens worldwide? And what can the people in this room do to ensure the future of first-hand, fact-based reporting that’s essential to the health of our democracies?

“I fear we are careening toward a future with fewer and fewer journalists to do the expensive, difficult work of original reporting—going to places, talking to people, digging up information, covering important issues and events, providing context and analysis, investigating the powerful. A future where a crucial wellspring of a healthy society and a stable democracy—the truth, understanding and accountability provided by original journalism—continues to dry up.”

Governments have frequently regarded the truth as an inconvenience in asserting claims about nuclear weapons, capitalising on the ease with which the public can be deceived about an unfamiliar and powerful subject.

After the 1945 atomic bombings in Japan, a handful of intrepid journalists defied US government claims to the contrary and offered first-hand reports—based on so-called shoe leather reporting—about the deadly effects of radiation at Hiroshima and Nagasaki.

In the run-up to the 2003 Iraq war, journalists relied on official sources, and—lacking the understanding with which to challenge false claims about Iraq’s possession of nuclear and other weapons of mass destruction—failed their audiences in their reporting.

Since that era, disinformation and misinformation have become epidemic. There are fewer reporters because the news industry has been diminished by loss of advertising revenue to social media and search engines and changing news consumer habits. At the same time, press freedom is under attack globally.

The tech giants building AI (six of the industry’s major players have been valued at $11 trillion) are fueling a continued decline in original reporting, Sulzberger said, “a continued surge in misinformation, propaganda, conspiracy theories, deepfakes and computer-generated slop. A public that continues to be radicalized by algorithms that amplify fear, anger and division.”

Although more people are turning to AI as a source for news, research by the BBC and the European Broadcasting Union said AI failed in terms of reliability, “falling short on basic criteria like accuracy and providing adequate context, things which are essential editorial values for quality journalism and news content. Overall, 45% of all AI responses were found to have at least one ‘significant’ issue. When including ‘some issues’, 81% of responses (were).”

The EBU wants publishers to have greater control over whether and how their material is used by AI. And where it is authorised for use, “clearer attribution and citations in an agreed format with prominent links to original material,” are needed, the EBU contends.

Or, as Sulzberger has suggested, “The clearest path to support quality reporting will be through direct relationships with audiences.”

In the AI storm, there may be some solace in the findings of a study by the Reuters Institute. “When the conversation turns specifically to journalism, public sentiment cools considerably,” writes Felix Simon, one of the report’s three authors. “Our report identifies a clear and persistent ‘comfort gap’ between human- and AI-led news production. Across countries, only 12% of respondents are comfortable with news made entirely by AI, rising slightly to 21% if there is a ‘human in the loop.’ Comfort, however, jumps to 43% for news led by a human with some AI assistance and 62% for news produced entirely by a human journalist.

“Americans are not outliers here. Their comfort levels mirror these global averages, showing a deep-seated preference for human oversight and authorship,” the report says. That preference alone will not preserve quality journalism in the AI age. Laws aimed at regulating the developers of AI models and compensating news organizations for use of their content will be required in countries around the world.