Seven people, not just one departure
Anthropic's founding group left OpenAI together. Dario Amodei, OpenAI's vice president of research, and his sister Daniela Amodei, also a vice president there, were joined by five more departing colleagues: Chris Olah, Jack Clark, Sam McCandlish, Jared Kaplan and Tom Brown, for seven co-founders total. [2][1]
TechCrunch's report, dated May 28, 2021, put the new company's seed funding at $124 million, led by Skype co-founder Jaan Tallinn with investors including Dustin Moskovitz and Eric Schmidt. Dario Amodei told the outlet Anthropic's goal was to make the research advances needed for more capable, general and reliable AI systems, then deploy them in a way that benefits people. [1]
The research they had already done together
Four of the seven, Amodei, Kaplan, McCandlish and Brown, had co-authored Scaling Laws for Neural Language Models at OpenAI in January 2020, the paper that showed language model loss falls as a predictable power law with more parameters, data and compute. [3]
Atlas interpretation: That result is the quiet premise behind Anthropic's founding thesis. If capability climbs predictably with scale, the people who plotted that curve had reason to expect more capable systems on a schedule, not a surprise, and reason to think safety research needed to start before the next jump. Anthropic's own account of its first year describes that work as making large models more steerable, interpretable and robust, not a plan to slow scaling down. [3][4]
A diplomatic account of an undiplomatic decision
Two years later, Dario and Daniela Amodei were still careful about how they described the break. Daniela said only that the group had 'a different vision for building safety into models from the beginning.' Dario framed Anthropic's role as competitive pressure applied gently: 'I think our existence in the ecosystem hopefully causes other organizations to become more like us.' [5]
Atlas interpretation: Neither sentence names a specific dispute inside OpenAI, and none of the sources here catch one on the record. What the 2021 and 2023 remarks agree on is the shape of the disagreement: not a break over whether AI should be built, but over how much safety work should be built in before scaling further, and by whom. [5][1]
The filing predates the funding
Anthropic, PBC was registered in Delaware on January 26, 2021, four months before the TechCrunch story that made it public. Anthropic's own account of that year says the company was founded at the beginning of 2021, spending its first months on interpretability and steerability research before the funding, and the news, arrived. [6][4]
Atlas interpretation: The event date here marks the public reveal, the moment the company and its funding became visible, which fits this dataset's rule of dating a commitment to when it became news. It just means the launch was closer to a coming-out than a beginning: the seven had already been working together for months. [6]
Sources
- Anthropic is the new AI research outfit from OpenAI's Dario Amodei, and it has $124M to burn
TechCrunch · May 28, 2021
- Daniela Amodei
Forbes · Sep 8, 2026
- Scaling Laws for Neural Language Models
arXiv · Jan 23, 2020
- Anthropic Raises Series B to Build Safe, Reliable AI
Anthropic · Apr 29, 2022
- Dario and Daniela Amodei
TIME · Sep 7, 2023
- ANTHROPIC, PBC - LEI: 984500B6DEB8CEBC4Z70
Bloomberg · Sep 8, 2026