Karen Hao probes OpenAI culture

OpenAI usually gets covered like a product company. Bigger models. More users. New safety tests. But Karen Hao is pushing a different frame — that the real story is also about culture, belief, and power inside the organization building the tools. Her May 10 interview, tied to her book *Empire of AI*, makes the case that OpenAI’s internal mythology around AGI helped shape the company’s behavior as much as any engineering roadmap. ### What’s actually new here? The new thing is not a fresh OpenAI announcement. It’s the way Hao’s reporting is landing now, through a widely shared interview that turns her years of reporting into a simple argument people can grab onto: OpenAI’s governance problem is cultural before it is procedural. In the video, she talks about “religious-like” dynamics, “machine god” mythology, and retreat rituals as part of a broader belief system around AGI. (youtube.com) ### Why does “cult-like” language matter? Because it changes the risk model. If a company is just overhyped, you worry about bad forecasts and wasted money. If a company starts acting like a belief community organized around a transcendent mission, dissent gets harder, internal tradeoffs get fuzzier, and ordinary guardrails can start to look like betrayal. That does not prove OpenAI is literally a cult. But Hao’s point is that the social structure starts to matter in the same way the model architecture does. (youtube.com) ### What is Hao basing this on? A lot of reporting. The preview for *Empire of AI* says the book draws on more than 150 interviews, including over 90 current or former OpenAI executives and employees, plus contractors with access to documentation about parts of model development practices. That scale matters because Hao is not making a vibes-only critique from the outside. She is stitching together a long internal history. (youtube.com) ### What picture of OpenAI emerges? Basically, a company that started with a nonprofit, safety-first identity and then got pulled into a resource arms race it could not resist. Hao’s own site frames the story as one of mission drift — OpenAI was supposed to check mercenary AI development, then became one of the main engines of the race, backed by Microsoft’s money and massive compute demands. The tension is not subtle: public-benefit language on the outside, empire-scale incentives on the inside. (s1.papyruspub.com) ### Why is AGI the load-bearing idea? Because AGI is vague enough to hold almost anything. Hao argues that even inside the field, people do not agree on what AGI really means. That makes it powerful as ideology. A fuzzy destination can justify a lot of concrete decisions — secrecy, urgency, fundraising, centralization, and the claim that normal rules should bend because the mission is historic. It works a bit like a moving finish line that still keeps everyone sprinting. (karendhao.com) ### Is this only about OpenAI? Not really. The interview uses OpenAI as the clearest case, but the broader target is the frontier AI industry. Hao says the industry has taken on a belief-system quality, and her book description widens the frame further — chips, data extraction, low-wage labor, energy, and water all become part of the same imperial supply chain. OpenAI is the focal company because it made the AGI story mainstream through ChatGPT. (youtube.com) ### So what should readers take from this? The useful shift is simple. Don’t ask only whether a lab’s model is safe. Ask whether the institution can tolerate criticism, define its goals clearly, and resist turning its mission into a moral blank check. Hao’s reporting matters because it treats OpenAI less like a dazzling machine shop and more like a power center with a theology. That is a harder story — but probably the more important one. (youtube.com)