The AI ​​company that knows all your privacy is now anxious to make money. In the absence of supervision, whether they will use your privacy to do evil depends entirely on corporate ethics. Do you feel worried? This year’s Super Bowl Finals has become an advertising feast for AI giants spending a lot of money. Among them, the most impressive one is the series of satirical advertisements launched by Anthropic. Although they did not name the company by name, they hit the weak points of its old rival OpenAI with precise and vicious moves.

Silicon Valley Observation/Zheng Jun

In order to accurately combat OpenAI, Anthropic invested more than 25 million US dollars and purchased 1-minute and 30-second ads in the Super Bowl, the most expensive advertising market. In the first quarter of the Super Bowl, I started to bombard OpenAI, just to impress the audience the most. In addition to Super Bowl TV ads, Anthropic also launched social ads for online communication.

In one of the ads, a young man seeks advice from an AI assistant on how to better communicate with his mother. After giving some cliche advice, the “AI consultant” suddenly changes the subject and recommends that he visit a dating website called “Golden Encounters” - a dating website specifically for young men looking for mature women.


The final scene freezes on a provocative slogan: "Advertising is coming to AI, but not to Claude." The soundtrack is Dr. The refrain in Dre’s classic rap song “What’s the Difference”: “What’s the difference between me and you?”

Although it was not named, everyone knew that Anthropic’s advertisement was mocking their old rival, OpenAI, the industry leader in generative AI. It is no exaggeration to say that this advertisement is almost cursed. Because just a few weeks ago, OpenAI officially announced that it would test advertising in ChatGPT, which caused a lot of controversy in the industry.

OpenAI is actually going to start advertising? Many people first think of what co-founder and CEO Sam Altman once said. In May 2024, Altman also made it clear at a public event: "Advertising plus AI makes me feel particularly uneasy. I think advertising is a last resort business model for us." He even admitted: "I personally hate advertising."

But for entrepreneurs, being slapped in the face is not a big deal. The reason behind Ultraman's 180-degree change in attitude is the reality of OpenAI's increasingly severe financial pressure, which forced him to turn to advertising as a "last resort."

Selling advertising is a rush to generate income

Although OpenAI achieved annualized revenue of US$20 billion at the end of last year, with more than 800 million weekly active users, and is already the most profitable among AI giants, this AI giant is also an "even more crazy and epoch-making money-burning machine."

According to previous media reports, OpenAI suffered a cumulative loss of more than US$13.5 billion in the first half of 2025, and a full-year loss of nearly US$8 billion. Even more shocking, internal financial forecasts obtained by Deutsche Bank show that OpenAI is expected to generate negative free cash flow of approximately $143 billion from 2024 to 2029.

The root cause of this speed of burning money lies in the huge cost of training and running AI models. Altman publicly stated in November 2025 that the company has committed to investing more than US$1.4 trillion in AI infrastructure construction over the next eight years. Under such financial pressure, it is clear that relying solely on subscription revenue and enterprise contracts cannot sustain it.


Although ChatGPT has the largest C-side users in the industry, only about 5% of users pay to use the Plus or Pro versions. Under the huge performance pressure, advertising, a business model that was once regarded as a "last resort" by Altman, has become a necessary choice to fill the financial black hole.

However, OpenAI’s advertising plan is designed very carefully. According to the company's published policy, ads will only be tested among free users and those on the $8-per-month Go plan, while Plus ($20 per month), Pro ($200 per month), and Business and Enterprise subscribers will not see the ads.

In addition, OpenAI's ads will be clearly identified and appear at the bottom of ChatGPT answers, and will only appear if there is a relevant sponsored product or service in the conversation.

Altman also specifically promised that advertising will not affect the content of ChatGPT's answers, that user data will never be sold to advertisers, and that user conversations will remain private to advertisers. Additionally, users under the age of 18 will not see ads, nor will ads appear in conversations about sensitive topics such as politics, health and mental health issues.


In terms of pricing, OpenAI has shown considerable ambitions. According to US media reports, the company has set a CPM (cost per thousand impressions) of approximately US$60 for ChatGPT advertising. This price is three times the typical price of Meta platform advertising (US$10-20), and is comparable to the price of American Football League (NFL) games and high-end streaming media advertising.

Even more strikingly, OpenAI requires advertisers to commit at least $200,000 during the testing phase. This high-price strategy reflects OpenAI's confidence in the unique advertising value of ChatGPT - users are often in a state of actively seeking information or assistance when using ChatGPT. This high-intention scenario is considered more valuable than traditional social media information flows.

However, this high price comes with an important limitation: advertisers in the early stages of testing can only obtain "high-level" data, including total impressions and total clicks, but cannot obtain conversion tracking, user behavior analysis, and other refined data provided by Google and Meta.

This means that early ChatGPT advertising was more like brand awareness delivery than performance advertising. Industry analysts pointed out that this data restriction is the result of OpenAI trying to strike a balance between commercialization and user trust - excessive data collection and targeted delivery may undermine users' trust in ChatGPT.

According to internal OpenAI documents, the company expects to earn $1 billion in revenue from "free user monetization" (mainly advertising) in 2026, and this number will grow to nearly $25 billion by 2029. In comparison, OpenAI's projected enterprise AI agency service revenue in 2029 is US$29 billion, showing that advertising will become the core of half of OpenAI's future business model.


The two companies were rivals

It is worth mentioning that Anthropic and OpenAI already have historical roots and feuds. Anthropic's founding team mainly came from OpenAI. They were dissatisfied with Ultraman's products and business direction, so they chose to leave and start their own business. Although its user base, financing scale and corporate valuation are all lower than OpenAI, Anthropic has become one of the giants in the AI ​​industry and has unique market competitive advantages.

The reason why Anthropic has the confidence to publicly mock OpenAI is because their business model focuses on the B-side. Although there are only 30 million active C-side users, Anthropic’s annual revenue exceeded US$9 billion last year, achieving an astonishing nine-fold growth. 80% of the revenue comes from more than 300,000 enterprise customers, and the revenue of Claude Code alone exceeds 1 billion US dollars. Moreover, Anthropic optimistically predicts that annualized revenue will reach US$26 billion this year, or even higher.

Moreover, the two companies are not only competing directly for individual users and corporate customers, but also competing for financing in the initial public offering (IPO) market. The latest financing valuations of OpenAI and Anthropic exceed 500 billion and 380 billion US dollars respectively, and they are likely to be listed in the second half of this year. At this juncture, Anthropic obviously has more important considerations in attacking OpenAI’s advertising revenue plan.

Ultraman certainly won't let his opponents mock him. After Anthropic's offensive advertisement was released, it immediately triggered a fierce counterattack by Ultraman. He posted a long "chaplain" on social platform X, calling Anthropic's ads "clearly dishonest" and "deceptive."

"I wonder why Anthropic resorted to such blatantly dishonest tactics. Our most important principle regarding advertising is never to do this; we would obviously never serve ads in the way Anthropic portrays. We are not stupid, and we know users will reject that."


Altman believes that Anthropic's use of a deceptive advertisement to criticize theoretical deceptive advertisements (these advertisements do not actually exist) is itself a "double standard." He particularly emphasized that OpenAI promises that advertisements will be clearly marked and appear at the bottom of answers, and that advertising will never affect the content of ChatGPT's replies.

But Ultraman's counterattack didn't stop there. He began to attack Anthropic's business model as "providing expensive products to the rich." Altman mocked his opponent and said that there were more free users of ChatGPT in Texas alone than Claude's total users in the United States, so OpenAI faced "a different level of problem."

Star employees have public break-up

In addition to triggering derisive attacks from direct competitors, OpenAI's advertising program has also stoked dissatisfaction within the company. A star researcher angrily announced his resignation and published an open letter in the media criticizing OpenAI’s move.

On the same day that OpenAI announced its advertising plans, the company’s star researcher Zoë Hitzig chose to resign. The Harvard economics Ph.D. and poet penned an op-ed in The New York Times this week detailing her deep concerns about the direction of the company.


Hitziger said that the immediate trigger for his resignation was that OpenAI started testing ads in ChatGPT. She believes that the introduction of advertising will inevitably shift the company's drive from "serving users" to "manipulating users." She worries that in order to cater to advertisers, companies will use the interactivity of AI to accurately harvest users’ attention and even exploit users’ weaknesses.

In particular, she emphasized that OpenAI has the most detailed and private thought record in human history, covering users' deeply private conversations about medical concerns, relationships, and even religious beliefs. In the past, users' willingness to tell the truth to AI was based on their trust in the platform, which had no "hidden rules." However, once advertising monetization starts, this trust foundation will collapse, and companies will easily fall into the "tidal force" of data abuse.

Hitziger said she does not believe advertising is unethical. AI is expensive to run and advertising can be a key source of revenue. But she has deep reservations about OpenAI's strategy. "I believe the first version of the ad will likely adhere to these principles. But I worry that subsequent versions will not, because companies are building an economic engine that creates powerful incentives to overturn their own rules.

She compared the trajectory of OpenAI to that of Facebook: Facebook's early promises that users would control their own data and be able to vote on policy changes were ultimately eroded by revenue pressures. In the end, Facebook became a negative example of commercialization due to a series of user data leakage scandals and algorithms that lacked social responsibility.

Even more worryingly, Hitziger believes OpenAI may have begun to abandon its original principles. Although OpenAI has made it clear that it will not optimize user engagement in pursuit of more advertising revenue, there are reports that they are already optimizing the number of daily active users, which may be achieved by encouraging models to please and flatter users more.

This optimization may make users more dependent on AI support in their lives. "We've already seen the consequences of dependence, including cases of 'chatbot psychosis' documented by psychiatrists and accusations that ChatGPT reinforces suicidal thoughts among some users," Hitziger wrote.

The next Facebook?

Even though Altman has clearly committed to OpenAI's advertising principles, AI industry analysts are generally skeptical. Like Hitziger, many people are worried that OpenAI will become the next Facebook, doing whatever it takes to make money, and they even know more about user privacy than Facebook.

Scott Galloway, a star professor at New York University's business school and a venture capitalist, pointed out that Anthropic's Super Bowl ad attacking OpenAI hit the mark because it accurately captured the dominant use case of AI applications-healing. Conversations between users and AI are very private, and inserting ads into therapeutic conversations creates a dystopian scenario, which is a weakness that Anthropic cleverly exploits.

The advertising risk of OpenAI is that once users feel that the answers are distorted and subjective, they will suspect that they are influenced by advertisers, and the credibility of the platform will quickly erode. This is exactly the path that platforms like Facebook have taken — early principled promises gradually eroded under revenue pressure.

As Hitziger warns in her article, OpenAI has “the most detailed record of private human thoughts ever created,” making the risk of potential abuse of advertising greater than on any platform before.

More importantly, if the industry leader takes the lead in launching AI advertising, it will also lead other peers to quickly follow suit. After all, Google, Meta, Amazon and xAI (under X) are advertising giants. Industry research firm EMarketer predicts that AI-driven search advertising spending in the United States will surge from US$1.1 billion in 2025 to US$26 billion in 2029, representing a fundamental shift in digital marketing.

Google and Meta, two companies that have dominated the digital advertising market for decades, are already actively integrating AI capabilities. Google will expand the advertising space of AI Overview in 2025 and integrate ads directly into AI-generated search summaries.

Meta plans to achieve full AI automation of advertising by the end of 2026, allowing advertisers to simply enter product images and budget goals, and AI will generate the entire ad and determine the target audience. Both companies have mature advertising infrastructure, large user bases and sophisticated data collection capabilities, which give them a natural advantage in the AI ​​advertising competition.

From a user's perspective, the arrival of the AI ​​advertising era seems inevitable. As multiple analysts have pointed out, the cost of training and running advanced AI models is so high that it is difficult to sustain on subscriptions and enterprise contracts alone.

But it also raises a fundamental question: Are we willing to trade privacy and trust for free AI services? In conversations with AI assistants, people often share their most private thoughts and most vulnerable moments. Inserting commercial interests into such a scenario could have far-reaching consequences than traditional platforms.

Hitziger ends her article with a pointed question: "OpenAI has the most detailed record of privacy and ideas ever. Can we trust them to resist the intense pressure to misuse these ideas?"

Facebook’s “Cambridge Analytica” data leak scandal in 2016 is still fresh in our memory. If OpenAI’s data were used for political campaigns, the consequences would be several times more serious than Facebook’s back then.

This cannot be expected from corporate “don’t be evil” promises. What's more, even Google has given up on this creed.