The Atlantic and OpenAI · Prep

The Professor's Questions

AI and the Ideas Economy. In March 2024, The Atlantic announced a strategic content and product partnership with artificial intelligence giant OpenAI. OpenAI would license The Atlantic’s content to train its models and respond to user queries, while The Atlantic would receive a fee, privileged access to OpenAI’s technology, and “premium” positioning within the tech giant. At the time, over 70 similar deals had been struck between publishers and AI companies. The Atlantic’s journalists, however, were incensed, concerned that partnering with OpenAI was “a devil’s bargain.” The New York Times had also just sued the tech giant for allegedly using 16 million pieces of the Times’ copyrighted content without permission to train its large language models. For Laurene Powell Jobs, the billionaire owner of The Atlantic, was it in the magazine’s best financial interest to continue the OpenAI partnership? What impact might a continued partnership have on the magazine’s journalism, or on the ideas landscape as a whole? What responsibility did Powell Jobs have – if any – to protect creativity and expression? Was Jobs undermining The Atlantic’s reputation, or was she bringing a new business model to bear on an industry facing major headwinds?

What is the ideas economy? In what ways does it impact the world in which a business leader operates?
Why did Laurene Powell Jobs and her Emerson Collective buy The Atlantic?
Should The Atlantic have partnered with OpenAI? Why or Why not?
What impact might a continued partnership with OpenAI have on the magazine’s journalism, and on the ideas landscape more broadly?

My Answers

Q1. The ideas economy

The ideas economy is the market that produces, distributes, and prices original thought — journalism, literature, science, argument, design, and the other things a society uses to decide what is true, what is worth paying for, and what is allowed.

It has three layers. Production is the scarce human work: a 168-year magazine, a novel that took three decades, a Pulitzer investigation that can force a resignation or free a wrongly convicted man. Distribution used to be printing, then the open web, then a search box. It is now also a chatbot that answers without sending the reader to the source. Capture is who gets paid. For most of The Atlantic’s life, capture was weak: culturally indispensable, rarely profitable, losing $4 to $8 million a year even after Bradley had put in more than $100 million. The 2024 P&L finally worked — $100 million of revenue, nearly two-thirds from 1 million paying subscribers — just as a new buyer of ideas appeared: model trainers who treat the archive as fuel.

This is the world a business leader already operates in, with or without a media P&L.

First, leaders live inside other people’s sentences. Regulation, talent, customers, and permission to operate are downstream of what magazines, courts, and now models say is legitimate. Powell Jobs bought a magazine because she wanted a voice in that layer, not because the 1990s P&L was pretty.

Second, every firm is now both a consumer and a supplier of ideas. Consultants using ChatGPT beat those who do not, but the gains sit mostly with low-baseline people and the variance across a team collapses. A 2025 MIT study in the case has ChatGPT users at the lowest brain engagement, sliding toward copy-and-paste. The leader who treats the model as a free intern is quietly thinning the firm’s own judgment.

Third, the legal design of this market is unfinished, and it is global. U.S. fair use has four factors and no formula. One Delaware judge said training a competing product on Thomson-Reuters headnotes was not transformative. Another judge said Anthropic’s use of seven million books was. Anthropic still paid $1.5 billion, about $3,000 a title. OpenAI’s own argument is national: training is “critical for U.S. competitiveness.” A 2022 AAAS study found text-and-data mining more common in countries with weaker copyright control. So a CEO in Boston, Shenzhen, or Singapore is already choosing a copyright regime when she chooses a model vendor.

Fourth, platforms have become the clearinghouse. Google’s AI Overviews cut referral clicks by more than 34 percent. Chatbots have replaced search for more than 25 percent of Americans. The FTC’s 2023 warning in the case is the industrial-organization sentence: firms that control the data may “wield outsized influence over a significant swath of economic activity.” That is not a media problem. It is a market-structure problem that happens to show up first in sentences.

The Global Markets point is simple. Ideas used to cross borders as magazines, books, and graduates. They now also cross as tokens in a training run. Whoever sets the price of those tokens — license, lawsuit, or “fair use” at $0 — is writing the trade rule for the next decade of knowledge work.

Q2. Why they bought The Atlantic

They bought a civic instrument that had finally learned how to lose less money, not a growth asset. The sentence Powell Jobs used in 2017 is the reason: a “critical mission at this critical time” — equality, the American idea, American culture and literature, and “our marvelous and sometimes messy democratic experiment.”

Three facts make this more than a press release.

One, the magazine had never been a financial prize. Bradley paid $10 million in 1999, put in more than $100 million, and still lost $4 to $8 million a year until a digital and events turnaround produced $1.8 million of profit in 2010. He sold because his adult children wanted nothing to do with it. A 70 percent stake for more than $100 million, paid by a woman then worth about $18.2 billion, is a stewardship check, not a PE entry multiple. Emerson is an LLC on purpose: venture, activism, and philanthropy at one table. It had already put money into Gimlet and Axios “for serious journalism for its own sake,” Bradley’s phrase. The name is Ralph Waldo Emerson, who opens the case: the poet and inventor who “unlocks our chains.” She said Emerson invites us “to transcend limitations that are placed on us by society.” Buying the magazine his circle founded is the most literal way to fund that sentence.

Two, she did not want to be a distant check-writer. A director at an Emerson-backed group said a philanthropist at her level can operate from a comfortable distance. She wanted to be “an active participant.” She added more than 100 people, about 30 percent of headcount, put in a paywall, rebuilt the typeface and nameplate, and told Bradley the magazine should operate on the frontiers of technology — after he had said its gift was not that. The expansion made the P&L red again. She still hired Thompson in 2020 with a different order: dig the company out and make it self-sustaining. Mission without a going concern is a subsidy. A going concern without a mission is just another website. She has been trying to hold both.

Three, the Silicon Valley habit came with the check. She runs Emerson like a start-up and sends sector leads out to “disrupt” their spaces. She takes “tech tours.” The same woman who promised — and kept — editorial independence also sat in OpenAI’s world through AI start-ups and the Jony Ive partnership. That is not a contradiction she hid. It is the marriage the case names on page 1: tech titans and award-winning journalists, under one owner.

So the purchase is not explained by the 2010 $1.8 million, and not contradicted by the later $20 million hole. She bought the right to keep an American ideas institution alive, to keep her hands off the edit page, and to push the business side toward technology. The OpenAI deal is where those three wishes stop being able to live in separate rooms.

Q3. Should they have partnered

Yes. They should have signed. They should renew. They should not renew the same deal.

The first deal was a price discovery, not a marriage. Thompson’s sentence is the right frame: the models “scraped without permission”; there are “a whole bunch of ways you can get paid for that”; you weigh what they offer against what a suit would cost and what it might return. That is not cynicism. That is the only adult arithmetic once the archive is already inside GPT-3.

The financial case for signing is clearer than the newsroom letter admits. The Atlantic in 2024 had $100 million of revenue and had only just climbed out of a $20 million hole. Exhibit 3’s visible prices run from the AP’s “single-digit millions a year” to News Corp’s $250 million over five years. The cluster that looks like The Atlantic — FT at $5 to $10 million a year, Axel Springer at $25 to $30 million over three years — is 5 to 10 percent of the magazine’s revenue. That is enough to fund most of the 2025 hiring wave: 30 editors at $200,000 to $300,000 is about $6 to $9 million. The lawsuit path is not free. The Times spent $10.8 million in year one, before any judgment. Statutory damages of $150,000 an article against a $10 license sound like a jackpot until you remember Anthropic got a mixed ruling and still settled books at about $3,000 a title, and until you remember OpenAI’s public line that any single source “didn’t meaningfully contribute.” A just-profitable magazine with one million subscribers should not be the test case that funds a decade of discovery.

The strategic case is the one Goldberg already made, accidentally. They moved off day-to-day news toward fewer, deeper stories. Commodity news is what a chatbot replaces first. A Pulitzer investigation that can oust a network chief, or free a prisoner, is what a model cannot originate and what a subscriber will still buy. Licensing the archive while thickening the newsroom is coherent: sell the past, raise the price of the next original piece. The later suit against Cohere — alleged full-article reuse and fake Atlantic bylines — shows they already know the difference between a paying counterparty and a parasite. That hybrid is the strategy, not a wobble.

The “no” side is real, and it should write the next contract, not kill the relationship.

Beres is right that OpenAI’s history is messy and that hallucinations at nearly 30 percent can smear a brand the magazine just spent years making worth $67 a subscriber. Lessin is right that a cheap license can fund the product that keeps the reader from ever arriving. Google Overviews already took more than 34 percent of referral clicks; chatbots already replaced search for more than a quarter of Americans. If “premium source” becomes a substitute for the site, Thompson’s “potential traffic source” is a story he tells himself. Powell Jobs’s silence is not editorial virtue on this file. Emerson’s AI book and the Ive–OpenAI hardware deal mean she must be able to show that renewal terms were set as if she had no other chip on the table.

So renew, on terms a seller writes:

Split the rights. Training is one license. Retrieval, quotation, and answer-generation are another. Do not sell both as a bundle at FT prices.
No verbatim regurgitation, and working citations only. Hallucinated links were the newsroom’s first exhibit. Technical controls belong in the contract, not in a blog post.
A most-favored reset. If the Times judgment or the next Anthropic-scale settlement sets a higher floor, this deal reopens. $3,000 a book is now a public number.
A traffic cap in the opposite direction. Never let AI referral become a material acquisition channel. The magazine already learned this with programmatic ads; it should not relearn it with chatbots.
Keep suing the ones who do not pay. Cohere is the deterrent that makes the OpenAI check mean something for the rest of the ideas market.
Keep the owner’s other bets off the term sheet. Recuse, disclose, or put an independent director on the renewal. The newsroom cannot be asked to trust a deal the owner also sits on the other side of.

The cost of this package is a smaller check, a slower product roadmap, and a colder relationship with a lab that likes to move fast. The cost of a blind renewal is the sentence Lessin already wrote, printed in The Atlantic, about The Atlantic. The cost of walking away is to spend like the Times without the Times’ balance sheet, while the archive stays in the model and the hiring spree is paid for by subscribers alone. Powell Jobs bought the magazine to defend an American idea. Paying writers $200,000 to $300,000 with money the models now have to acknowledge is a defense. Giving the models a quiet, perpetual feed is not.

Q4. Impact on journalism and the ideas landscape

A continued partnership, on seller’s terms, buys time for the newsroom and sets a price for the rest of the market. A continued partnership on the first contract’s terms subsidizes replacement and flattens the ideas landscape. The variable is the contract, not the logo on the press release.

On journalism, the near-term effect is already visible and mostly good. The magazine is hiring while peers cut. Thirty editors at $200,000 to $300,000 in a single half-year is not “billionaire benevolence” as a mood. It is a bet that human reporting is the scarce input — the same bet the case floats at the end when it asks whether AI labs need writers to avoid model collapse. Goldberg’s shift to fewer, deeper stories is the right editorial answer to a chatbot that can summarize yesterday. Three Pulitzers, a cover that can exonerate a man, a story that can move a network chief: that is the product a subscriber still cannot get from a 30 percent hallucination engine.

The medium-term effect is the one the newsroom fears, and it is not imaginary. If ChatGPT answers with a polished paraphrase plus a dead link, the subscriber never arrives. Overviews at minus 34 percent of clicks, and chatbots at more than 25 percent of U.S. search, are already that world. Incentives inside the magazine will tilt, quietly, toward being cited by the model: shorter explainers, more evergreen explainers, fewer expensive investigations that a model cannot use cleanly. That is how a hiring spree becomes a content farm with better furniture. Brand risk sits on the same page. A hallucinated “Atlantic” claim is not a traffic problem. It is a defamation problem wearing the magazine’s typeface — which is why the Cohere suit matters as much as the OpenAI check.

On the ideas landscape, scale is the tell. By April 2025 OpenAI had twenty publisher partnerships covering more than 160 outlets; about seventy-two licenses existed across AI firms. That sounds like a market. It is a market only for people who can hire Nick Thompson. Min Jin Lee’s Instagram note in the case is the other half: three decades, a legal career left on the table, and then a model that “stole my stories.” If the going rate is a low-seven or mid-seven figure check to a masthead, and $0 to everyone else, two things follow. First, the training corpus concentrates around a few English-language institutions that can litigate or bargain. Second, the next Pachinko does not get written, or it gets written by someone who already has a trust fund. The FTC’s worry about outsized influence is this, in prose. Adobe’s Content Credentials are a useful patch for attribution. They do not create a living.

Creativity research in the case is the third effect, and it is the one business leaders will feel even if they never buy a magazine. AI lifts the floor and sands the ceiling. Divergent-thinking scores beat all but the best humans. Stories written with ChatGPT were more useful for weak writers and more like each other as a set. Consultants got better together and more alike. MIT’s ChatGPT users got lazier. Kaufman and Beghetto’s Four-C model (Exhibit 4) is the vocabulary: mini-c and little-c everyday work will be automated and homogenized; Big-C and Pro-C work — the steam engine, the choreopoem, the investigation that takes a year — become more valuable and harder to fund if the distribution layer no longer sends a reader, or a royalty, back. A partnership that only dumps The Atlantic into the blender accelerates that flattening. A partnership that forces citation, forbids regurgitation, and recycles the fee into $300,000 reporters is one of the few ways a private owner can put sand in the gears.

Powell Jobs’s responsibility is not to freeze the technology. Ive already said the quiet part: if you make something new, some consequences will be harmful, and responsibility is “a determination to try to be useful.” Her version of useful, given why she bought the magazine, is narrower than OpenAI’s. She does not owe Altman a cheap corpus. She owes the next Emerson a world in which a new thought still has a payer. Renewing without terms fails that. Walking away and letting fair use stay at $0 for everyone smaller than the Times also fails that. The owner who sits in both rooms has to set a price that other rooms can copy.

Case Summary

In March 2024, The Atlantic announced a two-year content and product partnership with OpenAI. OpenAI got a licence to use the magazine’s content to train its models and answer user queries; the magazine got a licensing fee, privileged access to OpenAI’s technology, and a place inside the tech giant as a “premium news source.” What CEO Nicholas Thompson cared about wasn’t only the money, but “a potential traffic channel” and the chance to “help shape the future of AI.” The newsroom exploded. Journalists wrote an open letter to Thompson and editor-in-chief Jeffrey Goldberg: AI cannot replace journalists, and no outlet that still wants to live up to this magazine’s century-and-a-half of standards can allow that to happen. Editor Damon Beres called it “a devil’s bargain” in the magazine’s own pages. Jessica Lessin, founder of The Information, also wrote in The Atlantic: these deals amount to settling without suing, trading hard-won credibility for a bit of cash, while the other side undervalues you and builds products plainly designed to replace you.

From day one, this magazine was more an American idea than a good business. In 1857, a group of abolitionist intellectuals — Emerson, Longfellow, Holmes — founded The Atlantic Monthly in Boston to talk about slavery, and about the tensions pushing the country toward civil war. It went on to publish Hemingway, Einstein and Booker T. Washington, as well as early work by Mark Twain, Henry James and Louisa May Alcott. Former editor Cullen Murphy put it this way: it advanced something you might call “the American idea.” Culturally indispensable, it rarely won on the books. In the 1990s it could lose as much as US$500,000 a year. In 1999, David Bradley — owner of National Journal and HBS MBA 1977 — bought it for US$10 million, and reportedly poured in more than US$100 million afterwards: better paper, dinners with advertisers, new subscription prices. The magazine still lost US$4–8 million a year. In 2004 “Monthly” came off the cover. Only when Bradley started running it like a startup — digital advertising rose from 9% to 45%, it launched city coverage and a digital edition, and the business expanded into events and marketing consulting — did it turn a US$1.8 million profit in 2010. His children didn’t want the magazine. He began looking for an exit.

In 2017, Laurene Powell Jobs, Steve Jobs’s widow, bought 70% through the Emerson Collective for more than US$100 million. Her net worth at the time was around US$18.2 billion. Bradley kept a minority stake and stayed on to run it for another three to five years. He told staff that Powell Jobs had always been at the very top of his list: Emerson was already investing in serious journalism “for journalism’s sake.” She described the magazine as having “a critical mission at a critical time” — fostering equality for all, articulating and defending the American idea, celebrating American culture and literature, and covering this wonderful and often messy experiment in democracy. The collective is named after Emerson. She said Emerson invites people to move beyond the limits society places on them. The organisation itself, though, is pure Silicon Valley: an LLC rather than a traditional foundation, putting venture capital, activism and philanthropy at the same table; it invested in Gimlet and Axios; it helped Iñárritu make a VR piece on immigration; its climate action and College Track work run at billion-dollar scale. She encouraged the heads of each area to “disrupt” their fields, and regularly made her own Silicon Valley “tech tours.” She rarely speaks publicly and prefers to give anonymously, but she is willing to get her hands dirty rather than write cheques from a comfortable distance.

After taking over, she promised not to touch the editorial side. Bradley said that line went down “very well” in the newsroom — the question everyone was asking then was whether the magazine would acquire a party line on “her issues.” At the same time she hired heavily: headcount grew by more than 100, about 30%; a paywall went up; design talent was brought in to redo the logo, create a custom typeface and simplify the layout; and she pushed the whole company to use technology more deeply. Bradley’s words at the time: The Atlantic’s core competence is not operating at the frontier of technology. Her reply: it should be. The expansion was expensive. Profits of roughly US$10 million a year when she arrived turned back into losses. Ta-Nehisi Coates and other star writers left. The conservative writer Kevin Williamson was hired and fired in quick succession, rattling the newsroom. When the pandemic hit in March 2020, subscriptions rose, yet the magazine still laid off 17% of its staff. For those who had hoped “billionaire benevolence” could shield it from the industry’s headwinds, the layoffs were a sign that the inevitable obstacles had already arrived.

At the end of 2020, former Wired editor-in-chief Nick Thompson became CEO. Powell Jobs’s brief to him: dig the company out of its hole and make it a business that could support itself. He inherited a US$20 million deficit, and traffic was sliding as the pandemic eased. He immediately raised subscription prices by 50% and launched a dynamic paywall that targets readers by their likelihood to convert. Editor-in-chief Goldberg shifted the magazine away from daily news toward fewer, more deeply reported pieces. He said that Powell Jobs and the team all believed there was still a large, untapped audience willing to pay for the highest-quality journalism. By 2024 the magazine was profitable again, with more than 1 million subscribers and US$100 million in revenue, nearly two-thirds of it from subscriptions — up from less than half before. What advertising remained moved from programmatic toward events and direct brand relationships. The editorial scorecard looked just as good: three Pulitzers under Thompson, two of them won by Jennifer Senior and Caitlin Dickerson, freshly poached from the Times; the almost headline-free October 2024 cover won ASME’s best cover of 2025; a story on then-CNN chief Chris Licht led to his resignation, and a cover story helped exonerate a wrongly convicted man. Goldberg’s words: we treat it almost as an ideology, an idealism — what we make is worth paying for.

Over the same period, OpenAI turned the tech industry upside down. It started in 2015 as a non-profit backed by Musk, Thiel and Hoffman. Altman became CEO in 2018, and a capped-profit subsidiary was later added. ChatGPT launched in 2022, passed 30 million users in two months, and became one of the fastest-growing software products in history. Microsoft’s investment reached US$14 billion. Google, Meta and Amazon rushed to build the same thing. In 2023 the non-profit board tried to oust Altman and failed; a year later he dismantled the original non-profit structure. The models need enormous amounts of text. Convinced that scale buys performance directly, AI companies tested the limits of copyright and swallowed the internet whole: code from GitHub, homework from Quizlet, subtitles from YouTube, almost every English-language book, essay, poem and news story they could find. Meta executives even discussed buying Simon & Schuster purely for the data. Altman himself admitted that the usable data online would run out. One NIH-linked study found ChatGPT-4’s hallucination rate was close to 30%.

In December 2023, The New York Times sued OpenAI and Microsoft, the first major American media outlet to do so. The complaint said they had copied and used millions of the Times’s copyrighted works without permission to train their models; in Common Crawl, a “copy of the internet,” there were 16 million unique pieces of Times content alone. GPT-3, 3.5 and 4 were estimated to have been trained on 45 TB of data, roughly a 3.7-billion-page Word document. The Times sought billions of dollars in statutory and actual damages and demanded the destruction of models and data built with its copyrighted material. OpenAI mounted a fair-use defence and said that only weeks earlier, talks with the Times about a strategic partnership had been “progressing constructively”; it also wrote that any single source like the Times does not meaningfully contribute to the training of existing models and would not have sufficient impact on future training. It called training “fair to creators, necessary for innovators, and critical for US competitiveness.” Sarah Silverman, the Authors Guild (including Franzen and Grisham) and Getty (alleging Stability AI copied more than 12 million images) sued one after another. By May 2025 there were about 40 pending copyright cases. Condé Nast’s CEO once said content must not be stolen, distorted and trafficked — he later signed a deal with OpenAI himself. Frank Pine of Alden’s newspapers put it more bluntly: they pay their engineers, they pay for servers, they have astronomical valuations, but they don’t want to pay for the content without which there is no product. That isn’t fair use, and it isn’t fair.

There is no one-size-fits-all formula for fair use. Copyright law lists four factors: the purpose and character of the use (commercial or not, “transformative” or not), the nature of the work, the amount and substantiality of what is used, and the effect on the existing or potential market for the original. The Supreme Court once called the fourth “undoubtedly the single most important.” Experts are split. Stanford’s Lemley argues that models learn uncopyrightable ideas, facts and linguistic structures, not creative expression. Berkeley’s Samuelson argues courts have generally been lenient toward copying done to extract information and thereby advance the public interest. Loyola’s Hughes argues that if we accept models are “learning, training, memorising,” then that ingestion is quasi-expressive. The case law is split too. In February 2025, Delaware judge Bibas sided with the copyright holder in Thomson Reuters v. Ross Intelligence: using the other side’s lawyer-written headnotes to train a competing research product was not transformative, and the potential market was “obvious.” A few months later, Judge Alsup granted partial summary judgment in the Anthropic case: using 7 million books was transformative — “everyone reads books and then writes new text” — but Anthropic would still have to go to trial over alleged piracy for not having bought a single legitimate copy. In September 2025 Anthropic settled for US$1.5 billion, about US$3,000 per book, split between authors and publishers. It is the largest settlement in US copyright history. The Times, for its part, maintains that ChatGPT “regurgitates” articles verbatim, so readers no longer need to visit the Times; that Google’s AI summaries have already cut outbound clicks by more than 34%; and that chatbots have already replaced search for more than 25% of Americans. OpenAI counters that the Times manipulated prompts to coax out the original sentences, which is not how most users use it; it also says it helps journalists work more efficiently and sends readers back via links.

Thompson chose a different road from the Times. He acknowledged that the big models had “obviously scraped without permission.” Then he did the maths: there is a whole range of ways to get paid, so put what the other side offers on the same scale as the cost of suing and the damages you might win, and then choose. The exact terms were not disclosed. OpenAI was signing deals at a furious pace, and industry estimates of annual fees ranged from a few million to US$50 million (Exhibit 3: News Corp US$250 million over 5 years; FT US$5–10 million a year; Axel Springer US$25–30 million over 3 years). Thompson said the magazine would not become overly dependent on AI traffic; it would keep “telling stories in the best way possible” and then “distribute like crazy.” Another line of his: if content like The Atlantic’s goes into these models, the world will be a better place. The journalists weren’t buying it. They pointed out that ChatGPT hallucinates links, invents allegations, and steers readers to aggregator pages rather than the original site; they also worried about partnering with a company that, in litigation, was demanding to see all of Times reporters’ notes — which pushes right up against the limits of journalistic ethics. Powell Jobs never spoke publicly about the deal. She kept her promise of editorial non-interference: she didn’t meddle in planned candidate endorsements, nor did she bend editorial policy to be Trump-friendly; when Goldberg was mistakenly added to a classified Signal group where the Trump administration discussed war plans, she stood by her editor. But her interest in AI is public. Emerson has invested in several AI startups and played a key role in the roughly US$6.4 billion hardware tie-up between OpenAI and Jony Ive, a long-time friend of the Jobs family — the collective took part in the US$225 million round in Ive’s company before OpenAI acquired it, and had also backed Ive’s design firm LoveFrom. In 2025 she and Ive made a rare joint appearance: Ive talked about how innovation has unforeseen consequences, some wonderful, some harmful, and how he still feels responsible; she added — we now know unambiguously that some technologies have dark uses, but those are side effects.

The case stops at the table just before renewal. The news industry is preparing for “Google Zero”; Business Insider reportedly made layoffs because of falling traffic, and a former employee said it was “built for an internet that no longer exists”; some executives worry that AI partnership fees won’t make up for declining reader revenue. The wider question is the effect on thinking and creativity. Kaufman and Beghetto’s Four C model separates everyday little-c creativity from steam-engine-scale Big-C. The early research tells a tangled story: AI outperforms all but the very best humans at divergent thinking; using ChatGPT to help with storytelling raises novelty and usefulness for people with a low baseline, but the AI-assisted stories all look more alike; management consultants using AI outperformed those who didn’t, with the gains going mainly to low scorers and the variance in team output flattened; in a 2025 MIT study, ChatGPT users showed the lowest brain engagement, grew lazier with each essay over a few months, and by the end were often just copying and pasting. Jia Tolentino worries it makes people rely less and less on their inner resources. Natasha Lyonne, meanwhile, announced she would make a film using an “ethical” model not trained on copyrighted material. Under Powell Jobs, The Atlantic has not only returned to profit but put the money back into journalism: in the first half of 2025 it hired 30 editors at salaries of US$200,000–300,000. Most peers were cutting staff. Is this just one of the few self-sustaining outlets hiring by the old logic, or has she seen another layer of the AI era: the people building the models need writers? Trained only on synthetic data, models may collapse, and hallucination rates are already rising inexplicably. Is the partnership between writers and AI companies the key to both sides surviving economically, or is it accelerating journalism’s replacement and pushing creativity flatter? Powell Jobs has one foot in business and one in ideas. The case asks: what is the best way to steward this magazine through the AI era, and how will her decision reshape the wider business of ideas?

Key Data

Dimension	Figure / Fact	Source
Decision point	Two-year agreement nearing expiry; whether to renew and on what terms	p.1–2, p.10
Financial history	1990s annual losses up to $500K; bought for $10M in 1999, >$100M invested afterwards; annual losses $4–8M; 2010 profit $1.8M	p.2
2017 deal	70%, >$100M; Powell Jobs net worth ~$18.2B; Bradley keeps minority stake and runs it 3–5 more years	p.2
Post-acquisition expansion	+100 people (+30%); paywall; profit of ~$10M turns back to losses; 17% laid off in March 2020	p.3
Thompson era	$20M deficit at end of 2020; subscription price +50%; 2024 profitable, >1M subscribers, $100M revenue, nearly 2/3 from subscriptions; 3 Pulitzers	p.3–4
2025 hiring	30 editors in H1, $200–300K salaries	p.10
OpenAI scale	ChatGPT >30M users in two months; Microsoft up to $14B; trained on 45 TB ≈ 3.7B pages	p.4–5
Quality risk	ChatGPT-4 hallucination near 30%; AI summaries cut clicks 34%+; chatbots replace search for >25% of Americans	p.4, p.7
NYT case	16M pieces of Times content; statutory damages up to $150K/work vs typical licence $10/work; first-year law firm fees $10.8M	p.5, p.9
Industry structure	20 publishers / 160+ outlets; ~72 licensing deals; ~40 pending copyright cases	p.1, p.5
Licence price anchors	News Corp $250M/5 yrs; Dotdash $16M/yr; FT $5–10M/yr; Axel Springer $25–30M/3 yrs; range: a few million–$50M a year	Exhibit 3; p.7
Estimate for The Atlantic	Terms undisclosed. FT / Axel tier ≈ $5–10M/yr ≈ 5–10% of 2024 revenue	Estimate: 5–10 ÷ 100
Anthropic	$1.5B settlement, ~$3,000/book	p.6–7
Owner overlap	Emerson in Ive’s $225M round; OpenAI–Ive ~$6.4–6.5B	p.8
Subsequent stance	Feb 2025: The Atlantic sues Cohere, alleging it misappropriated full texts and passed off shoddy AI copy as the magazine’s	Endnote 131
Creativity research	AI raises the floor, flattens variance, lowers brain engagement; Four C model in Exhibit 4	p.9–10; Exhibit 4

Estimate: at FT’s $5–10M/year, this licence is worth roughly 5–10% of 2024 revenue, or roughly 50,000 to 100,000 annual subscriptions (spreading $100M × 2/3 ≈ $67M of subscription revenue across 1 million subscribers gives about $67 each; so $5M ÷ $67 ≈ 75,000 subscribers, $10M ÷ $67 ≈ 150,000 subscribers). It can’t buy a new magazine, but it could pay for a large share of that H1 2025 round of 30 well-paid editors (30 × $250K median ≈ $7.5M). The Times’s $10.8M first-year legal bill already exceeds one FT-tier licence. Litigation is a rich company’s game; The Atlantic has only just returned to profit.

Key People

Person	Role	Position / Role in the Story	Relationships
Laurene Powell Jobs	Founder of Emerson Collective, controlling shareholder of The Atlantic	Holds the decision. Bought the magazine for “the American idea”; promised not to touch editorial; has real-money interest in AI; ultimately owns the renewal question	Controls via the collective; investment overlap with Jony Ive / OpenAI; has not spoken publicly about the deal
Nicholas Thompson	CEO, former Wired editor-in-chief	Commercial driver. Admits the scraping was unauthorised, but uses licensing, technology and traffic to offset the cost of suing	Given the “self-sufficiency” brief by Powell Jobs; the newsroom’s open letter was aimed at him
Jeffrey Goldberg	Editor-in-chief	Shifted the magazine from daily news to fewer, deeper investigations; believes people will pay for the highest quality	Gatekeeper of editorial independence; backed by the owner in the Signal group affair
Newsroom journalists / Damon Beres	Internal opponents	“A devil’s bargain”; hallucinations, fake links, journalistic ethics, industry replacement	Open letter sent to Thompson and Goldberg
Jessica Lessin	Founder of The Information, writing in the magazine	These deals are settling without suing, trading credibility for cash, funding those who will replace you	Outside media figure using The Atlantic’s pages to oppose The Atlantic’s deal
Sam Altman / OpenAI	Counterparty	Fair use + “new revenue models for creators”; said the Times is “on the wrong side of history”	Signed with 20 publishers; simultaneously besieged by some 40 cases including the Times’s
The New York Times	External comparison	Sues for damages and destruction of models; can afford $10.8M in the first year	Same industry as The Atlantic, opposite strategy
David Bradley	Former controlling owner	Took the magazine from annual losses to a 2010 profit, then sold to someone he believed would pay “for journalism’s sake”	Kept a minority stake, still ran it during the transition
Jony Ive	Apple’s former design chief, io / LoveFrom	Hardware with OpenAI; speaks about technology’s dark uses and responsibility	Long-time Jobs family friend; Emerson is an investor
Judges Bibas / Alsup	External legal pressure	One ruled training a competing product is not fair use; the other ruled book-reading-style training is transformative	The law itself is split, giving the renewal negotiation both flexibility and risk

Decision rights: renewing, changing terms, or switching to litigation rests with the controlling shareholder, Powell Jobs. Day-to-day commercial design rests with Thompson. Editorial direction rests with Goldberg — the owner promised not to interfere. Emerson provides the money. Opposition comes from the newsroom and some peers. External pressure comes from the Times lawsuit, some 40 pending copyright cases, the clicks siphoned off by Google’s summaries, and her own investments in the OpenAI ecosystem.

Core Tensions

Whether to keep partnering with OpenAI (the professor’s Q3, put first): For renewal — the scraping has already happened; by peer benchmarks the licence fee is a few percentage points of annual revenue, enough to fund a round of well-paid editors; not signing won’t stop summaries and chatbots from siphoning off traffic (-34%, >25%); litigation costs $10.8M in the first year alone, and The Atlantic has only just become profitable. Against — Beres’s devil’s bargain, Lessin’s “funding your replacement,” hallucinations damaging the brand, the owner’s overlapping interests with OpenAI / Ive, open opposition from the newsroom.
Is fair use transformative learning, or taking a tradable market for free: Lemley / Samuelson / Alsup argue for “read books, then write new text”; Hughes / Bibas / the Times argue for “regurgitation + substituted clicks + a potential licensing market.” The fourth factor is key: once publishers start selling training rights, not buying them destroys that new market.
Powell Jobs bought a mission, but the books demand self-sufficiency: in 2017 the word was the American idea; in 2020 Thompson’s brief was to close a $20M deficit. Renewing only for short-term money would seem to betray the reason she bought the magazine; turning that short-term money into 30 editors at $200–300K each would seem to fulfil it.
The owner is also an AI investor: she promised not to touch editorial, and kept that promise. She made no promise to keep her commercial distance from OpenAI. Emerson’s AI portfolio and the Ive deal mean the claim that “renewal is good for the magazine” has to survive a conflict-of-interest check.
A deal can save a single magazine and still ruin the supply of ideas: big outlets can sign deals, while small creators get scraped for free (AAAS: countries with weaker copyright control produce more text-mining research; the FTC worries the giants could use control of data to sway large swathes of economic activity). If models stop paying for new human work, who will write the next book that takes thirty years to write (Min Jin Lee)?
AI raises the floor and flattens the ceiling: both the consultant and writing studies show low-baseline people benefit, outputs look more alike, and brain engagement falls. If renewal only turns The Atlantic into a “premium source” inside the model, the variance of the ideas market will keep being flattened.