Discurso do presidente do conselho do New York Times: Como a mídia deve sobreviver na era da IA? Eu aconselho a indústria a se levantar e resistir

The New York Times Chairman and Publisher A.G. Sulzberger delivered an opening speech at the WAN-IFRA World News Media Conference in Marseille on June 1, characterizing AI companies' use of news training models as "blatant intellectual property theft," revealing that The Times has spent over $20 million to sue OpenAI and Perplexity, and calling on the global media industry "to no longer remain silent."
(Background: CNN sues Perplexity for copyright infringement: 17,000 articles copied verbatim, bypassing paywalls)
(Additional context: Google Search undergoes the biggest transformation in history: repositioning Search as an AI-powered one-stop portal)

Table of Contents

Toggle

  • Four raw materials, three paid, one stolen
  • $11 trillion industry, 0.5% to content creators
  • One newspaper closes every three days
  • "Killed by its own copies"
  • "AI companies closer to Napster, not libraries"
  • Media too silent, too passive, too fragmented
  • News must be uniquely attractive

Key Highlights

  • A.G. Sulzberger accuses AI companies of "blatant intellectual property theft" at the Marseille WAN-IFRA conference, with The Times having spent over $20 million to sue OpenAI and Perplexity
  • Six major AI companies have a combined market value of $11 trillion, but less than 0.5% of their investments go to content creators; in the US, 75% of journalists have been lost over 20 years, with over 3,000 newspapers closed
  • Sulzberger calls for global media unity to promote legislation, require AI crawlers to identify themselves, and compares AI companies to "pirate music platform Napster"

Earlier this month, at the Marseille WAN-IFRA World News Media Conference in France, A.G. Sulzberger, Chairman and Publisher of The New York Times, did not exchange pleasantries with the audience. Instead, he directly addressed the global media industry with accusations against AI.

"Tech giants are strip-mining news websites without permission or compensation." ("Tech giants strip-mine news websites without permission or compensation.")

The speech's title was "AI, the News Industry, and the Uncertain Future of the Public Square." Over 40 minutes, Sulzberger dissected the issue very clearly. Here is his reasoning, in his own words.

Four raw materials, three paid, one stolen

Sulzberger breaks down the construction of AI models into four raw materials.

  • Talent (engineers)
  • Computing power (infrastructure)
  • Energy (electricity)
  • "Data"

He pauses notably on the word "data," because in the context of AI companies, "data" truly means books, movies, music, and news reports.

The first three raw materials are purchased at market-high prices. Engineers earn over a million dollars annually, data centers and electricity contracts cost billions. But for the fourth, they choose not to pay.

He quotes OpenAI's own words: "It is impossible to train state-of-the-art AI models without using copyrighted material."

Microsoft's vice president also admits: "High-quality content significantly improves response quality."

Sulzberger makes it clear: "What determines AI quality is not architecture, not hyperparameters, not optimization choices. It’s your dataset, and nothing else."

AI operators know news is valuable; they admit it, then take it directly.

$11 trillion industry, 0.5% to content creators

Sulzberger lays out the figures to highlight the contrast.

Six leading AI companies have a combined market value of $11 trillion, three times France's GDP. By 2025, private AI investments in the US will reach $350 billion, still accelerating in 2026. The global creative industry employs over 50 million people, with an annual output of about $12 trillion.

Yet, the compensation these AI companies provide to content creators is less than 0.5% of their investments.

He characterizes this as "blatant intellectual property theft on an unprecedented scale."

He then shifts focus to his own newspaper, noting that in 2025, The New York Times published nearly 500,000 articles, costing over $200 million. With over 70 reporters stationed in Ukraine, the paper has accumulated 20 million original articles over 175 years.

The Times is the largest proprietary content source in major AI training datasets.

One newspaper closes every three days

Next, a series of numbers that require no explanation.

Over the past twenty years, the US has lost 75% of its journalists. More than 3,000 newspapers have shut down, and now, every three days, another closes. Newspaper ad revenue has plummeted 80%. Meta’s ad revenue is eight times that of all newspapers combined.

AI search accelerates this depletion. "Getting a Google user to click a link now is ten times harder than ten years ago." Traffic to news sites from other AI models is 96% lower than from Google Search, and the largest newspapers have seen traffic drop over 45% amid AI competition.

Sulzberger says: "I worry we are heading toward a future with fewer and fewer journalists, fewer and fewer to produce those expensive, difficult original reports."

The decline of local news is not just a commercial issue; research shows that communities without local news see declining trust, reduced civic participation, increased isolation, and rising corruption. He calls this "a tragedy of the public sphere."

"Killed by its own copies"

Sulzberger quotes Canadian author Margaret Atwood to describe the current situation: "Killed by its own copies."

The irony of this metaphor is that the "copies" are often not even good enough.

Research by the European Broadcasting Union finds that nearly 50% of mainstream AI assistants give "seriously distorted" answers to news questions. Google and Apple make major errors when rewriting news headlines with AI. Perplexity's bot falsely claimed conservative commentator Charlie Kirk was assassinated, while Grok claimed he was still alive.

The most absurd detail is that when Microsoft launched its AI assistant Copilot, the disclaimer read: "For entertainment or reference purposes only. Do not rely on Copilot for important decisions."

A product claiming to replace news, yet telling users not to take it seriously.

"AI companies closer to Napster, not libraries"

Sulzberger chose a metaphor everyone understands.

"These companies adopt a parasitic stance, more akin to Napster, the old illegal music platform."

He then names them one by one.

OpenAI and Microsoft are the main targets of The Times' lawsuit, filed in December 2023, which has been ongoing for over two and a half years and cost more than $20 million. Most news organizations lack the resources for such a prolonged battle.

Meta trained its large language models on "a notorious illegal database of copyrighted books."

Perplexity "openly disregards" website crawling restrictions, using "hidden, unannounced crawlers" to bypass paywalls and distribute content.

Anthropic, often touted as a company committed to ethical AI, "refuses to pay for news."

Google increasingly answers user questions directly in search results rather than directing to original news sources. Even so, Google’s referral traffic remains the highest among all AI platforms, not because it does better, but because "competitors are worse."

Sulzberger concludes with a number: about 30% of AI crawlers violate explicit website access restrictions.

Media too silent, too passive, too fragmented

In the latter half of his speech, he shifts from accusations to calls for mobilization.

"Our industry is too silent, too passive, too fragmented in the face of AI revolution abuses."

He proposes four actions:

  • First: Uphold intellectual property rights and insist these rights be respected
  • Second: Carefully handle AI licensing agreements to ensure fair compensation and meaningful control
  • Third: Push for legislation to establish four layers of safeguards, including robust IP protections, mandatory crawler identification, transparency in news usage, and legal liability for AI-generated defamatory content
  • Fourth: Cross-industry unity, pooling resources, submitting amicus briefs, and collaborating with the film, music, and publishing industries

"We cannot let AI cheerleaders dominate the public discourse."

"We cannot stand by as AI companies attempt to permanently dismantle our rights to control."

News must be uniquely attractive

Sulzberger does not place all the blame on AI; he also issues a new-era demand to the media peers present.

He says over-reliance on platform traffic is a dead end; media must rely on themselves. "In a world mediated by AI, to become a destination, your news must be uniquely attractive."

He suggests four responsible yet bold actions: use AI responsibly but innovate boldly, build direct relationships with readers instead of relying on platforms, focus on original reporting rather than aggregation and clickbait, and explain to the public why news matters.

He also makes a distinction: "I am not saying AI, or the tech giants controlling this technology, are inherently evil or malicious."

"But rejecting a powerful new technology is a failed strategy."

The message is that the media should embrace AI advancements.

His final words are quiet, not angry, but a warning:

Even if everything seems fine now, remember, these early waves foreshadow an impending tsunami. We cannot be so naive this time. Information wants to be free, but it also wants to be expensive, because it is too valuable.

On average, every three days, a media outlet around the world permanently closes its doors.

Frequently Asked Questions

What does A.G. Sulzberger accuse AI companies of stealing?

Sulzberger accuses AI companies of using copyrighted news, books, music, and other content without permission or payment to train models, calling it "blatant intellectual property theft." The Times has spent over $20 million to sue OpenAI, Microsoft, and Perplexity.

How does Sulzberger suggest news media should respond to AI threats?

He proposes four actions: uphold intellectual property rights, carefully handle AI licensing agreements, promote legislation including mandatory crawler identification and transparency, and foster cross-industry collaboration. He also recommends focusing on creating "uniquely attractive" original content.

Ver original
Esta página pode conter conteúdo de terceiros, que é fornecido apenas para fins informativos (não para representações/garantias) e não deve ser considerada como um endosso de suas opiniões pela Gate nem como aconselhamento financeiro ou profissional. Consulte a Isenção de responsabilidade para obter detalhes.
  • Recompensa
  • Comentário
  • Repostar
  • Compartilhar
Comentário
Adicionar um comentário
Adicionar um comentário
Sem comentários
  • Fixado