Fact-Based Media in the Age of Accelerated Decay

On the bus leaving the seminar, I sat next to Edward Wong, a New York Times diplomatic correspondent and panelist in this year’s Milton Wolf seminar. Mr. Wong covered the Iraq War in the mid-2000s. That experience, he said as the bus rumbled west toward the hills outside Vienna, was formative for him. “I learned early that you can’t always trust the government,” he said. “Officials often have their own motivations and agenda," meaning those are not necessarily aligned with the public interest. As it happens, Mr. Wong was also a panelist at last year’s Milton Wolf seminar. I asked him how things were different this time.

Trump hadn’t been elected yet, he described to me, and there was more talk at last year’s seminar about the conflict in Gaza. The tone of this year’s seminar conversation, I noted, was dark. The seminar’s invitation asked us to consider the “seemingly endless decay wrought by late-stage capitalism,” a theme picked up by several of the panel’s speakers, who described, from both practitioners and policy-makers points of view, how public institutions were either failing to protect, or actively enfeebling, fact-based news organization worldwide. The metaphor of choice was the ouroboros, with each deterioration accelerating the deterioration of the others. How can these cycles be disrupted?

In this blog post as the 2025 seminar, that question won’t be (and wasn’t) answered. Instead I want to offer several reflections, expanding on Mr. Wong’s first effort, on the most distinctive challenges facing fact-based news organizations today, as distinct from previous years. What hasn’t the ouroboros seen before?

Authoritarianism on the homefront: no longer an anomaly

In the wake of the 2016 election, American fact-based media organizations engaged in soul-searching to understand how they, and many of America’s other civic institutions, could have been caught unawares by the rise of Donald Trump. Two theories emerged. The first was inward-looking: news media organizations had neglected America’s heartland, the “flyover states,” and their predominantly white working class communities, who together had propelled Donald Trump to victory (Carnes & Lupu, 2020, Morgan & Lee, 2018). The second was outward-looking. It sought to uncover foreign manipulation of America’s media, including both traditional and social media, through Wikileaks and Russian disinformation campaigns (DiResta et al., 2018). That investigation, with its substantial focus on Russian and other foreign interference, at least partially located the source of Donald Trump’s rise in a (digital) transgression of America’s borders and manipulation of its information environment. While the two theories were not mutually exclusive, they shared an implication that if the cause of the media’s failure could be diagnosed, its recurrence might therefore be prevented.

The 2024 reelection of Donald Trump revealed that 2016 was not an anomaly and the theories that explained authoritarianism’s American emergence were either incomplete or incompletely resolved. This year’s seminar did not advance new theories to explain its rise as much as document the ongoing accrual of damages in both America and globally, as well as put forward additional explanations for its recurrence. At the same time, many participants began to reconsider whether empowering government institutions to protect democracy or civil society could still be considered a viable approach, since state capture by authoritarian forces might weaponize those same newly-empowered institutions against democratic speech. This may have marked a substantial shift from previous years. With small-d democrats in power, as they had been since 2020 and otherwise more or less continuously in most American and European democracies since 1945, government seemed like a more viable protector of free media. Now, the case of Hungary felt less like an outlier than a harbinger: authoritarianism has taken root even inside America, and the acute political and commercial challenges faced by media organizations in Hungary are likely to spread.

Making matters more complicated for fact-based media, this authoritarian rise is occurring alongside an ongoing decline in journalism revenues worldwide (Hayes & Lawless 2018). During the seminar, speakers demonstrated how labor precarity, declining advertising revenues, and pinched reporting budgets have made the industry as a whole more vulnerable to authoritarian intervention. While some panelists highlighted success stories — they exist, including in hyper-competitive markets in Eastern Europe — the success stories are outnumbered by the news organizations dying or under threat, and face additional challenges due to the global erosion of public trust (Patterson & Urbanski, 2006). Several panelists urged self-sufficiency: for these outlets to survive, they must cultivate a reader base, and advertising base, that is also committed to their survival. While none of these observations are new, these preexisting trends are exacerbated by the return of authoritarianism even in the globe’s largest democracies.

Fact with fiction: AI hallucination and fake news

The American historian and philosopher of science, Thomas Kuhn, observed that the “answers you get depend on the questions you ask.” For much of the 2010s and 2020s, fact-based media organizations around the world asked the question: how do we restore truth to a place of authority within democratic discourse? In addition to be subtly backward-looking, for many, that task implied the struggle against mis- and disinformation in the wider news ecosystem. Papers were written, and are still written, about how better to inoculate the public against misinformation (Moore & Hancock, 2022), or how to flag and identify misinformation before it is widely circulated (Sharma et al., 2019). Those tasks of course remain an important tool in the preservation of democratic discourse worldwide.

The rise of large-language models (LLMs), however, and in particular their tendency to make things up or “hallucinate,” may be changing how we think about mis- and disinformation. Large language models have quickly become one of the world’s largest purveyors of incorrect information, but because that incorrect information has no “intention” in the conventional sense, is “algorithmic” in origin, many attendees hesitated to use “misinformation” when describing it (Berberette et al., 2024). Given that large-language models are too useful to abandon, as one attendee noted, the problem of hallucination will likely be here for a while. That may mean that we need to reconceptualize what misinformation is, and how best to combat it.

The other major development this year, also with its own ambiguous relationship with misinformation, is the rise of DeepSeek. Because the model is trained largely on Chinese‑language corpora scraped inside the Great Firewall, critics warn that it may inherit the normative frames and self‑censorship constraints embedded in those texts, and indeed early studies have shown that the system refuses to criticise the Chinese Communist Party or discuss sensitive events such as Tiananmen (Naseh et al., 2025). While this type of “imported bias” — bias via the unobserved data-gathering and fine-tuning processes — is not new, in the case of DeepSeek it also implicates the political objectives of a sovereign state, not just the now-quite-familiar biases from Silicon Valley’s lack of diversity or disguised commercial incentives (see, e.g. Turner, 2010). Yet because DeepSeek has outmaneuvered western top-of-the-line models like OpenAI by issuing a permissive license and performing well on multilingual benchmarks, many “third-party” companies and newsrooms (in particular in Europe) may spot an opportunity to diminish their dependence on Western models. Their enthusiasm underscores a dilemma: partnering with (or fine‑tuning) DeepSeek might lower costs and diversify model suppliers, but it also risks importing geopolitical misinformation embedded at the model‑weight level, which is harder to detect and root out.

No more going back?

Throughout the seminar’s discussions, one of the ongoing debates was how to identify our current moment within the scope of post-World War II history. On the one hand, it is always useful to revisit history to identify points of continuity with dilemmas faced by news organizations and other leaders in the past. On the other, if and when society is faced by a genuinely new problem, with little (or at least less) precedent in the recent past, continuing to look to history for guides risks repeatedly the mistake of Bertrand Russell’s famous Thanksgiving Day turkey, illustrating the problem of induction, who concluded that because he had been fed every day he will continue to be fed. The question for the seminar was to what extent the recurrence of authoritarianism in America and the proliferation of highly-capable reasoning models indicates a genuine break from the past, i.e. the arrival of Thanksgiving Day.

How one answers that question always seems (to me anyway) to implicate one’s own interests, demographics, and values. To a historian who has dedicated his life to understanding the past, or a wizened diplomat or journalist who has witnessed other moments that felt discontinuous but felt less so in retrospect, the present moment may appear more similar than different to past events. To the youth or the activist, the present moment may feel more like a sea-change, a new era, and indeed it profits those communities to speak about the present in that manner because doing so implicitly discounts the value of past experience and emphasizes the importance of action in the present moment. Personally, I mediate between these two perspectives by sometimes understanding human progress as neither linear (continuous) nor step-wise (filled with discontinuities) but something more like an exponential growth curve. Each era feels discontinuous because it represents an acceleration of human advancement over the era that came before it. In that way, both perspectives contain an element of truth: it is faster now, but then again, it always has been. I expect next year’s Wolf Seminar might feel the same.