These days, which company is the most successful in the artificial intelligence (AI) industry? When looking at technological achievements or the scale of investments, big tech companies like OpenAI, DeepMind, and Meta come to mind, but the place where the world's top AI talent is gathering is quite different. It is the AI startup "Anthropic."
According to the "2025 State of Talent Report" released on the 6th by the U.S. venture capital firm SignalFire, the rate at which personnel from OpenAI and DeepMind are transitioning to Anthropic is reportedly 8 times and 11 times higher than the reverse, respectively. Currently, the hiring pages of OpenAI and Anthropic each list approximately 300 and 200 job postings. This indicates that the "talent outflow trend" is essentially one-directional.
Anthropic's talent retention rate is at the industry's highest level. Of the personnel hired in the last two years, 80% are still employed, surpassing the figures of DeepMind (78%) and OpenAI (67%), as well as Meta (64%). It is establishing itself as a company where "once you enter, it's hard to leave."
Anthropic is a startup founded in 2021 by individuals who left OpenAI. All founding members, including co-founder Dario Amodei, come from OpenAI, and they have stated that they left the company because they felt concern over OpenAI's pace, which was moving "too quickly and too commercially." Founded with the core values of AI safety and public interest, Anthropic is organized as a public corporation.
The rush of transitions to Anthropic stems largely from a clash of philosophies rather than simply better compensation. Jan Leike, who led AI safety research at OpenAI, publicly criticized while transitioning to Anthropic last year, stating, "OpenAI became too focused on flashy products, and important safety research was pushed to the back." He also pointed out that "even the computing resources necessary for critical research were not sufficiently secured." Researchers from Google Brain and those formerly with DeepMind also joined Anthropic for similar reasons.
It has been reported that OpenAI and DeepMind are offering extraordinary compensation packages to retain key talent. OpenAI has reportedly proposed salaries exceeding $10 million (approximately 13.5 billion won) to some key researchers, along with a retention bonus of $2 million (approximately 270 million won) and stock compensation reaching $20 million (approximately 2.7 billion won) to prevent departures. DeepMind is offering some researchers paid leave for 6 to 12 months to block transitions to competitors. However, the case of Anthropic demonstrates that strategies based on "money" are no longer effective.
John Schulman, co-founder of OpenAI, also moved to Anthropic, saying he wanted to focus more on AI consensus, but later transitioned to a new startup led by Mira Murati, a former chief technology officer of OpenAI. Departures at the organizational level in the AI industry are no longer an exceptional phenomenon. Neil Horsky from DeepMind moved to Zurich, Switzerland, to establish Anthropic's European research institute, and security expert Nicolas Kallini, who worked at Google for seven years, chose to transition because he felt "Anthropic seriously addresses the AI security issues that I care about."
With the addition of numerous talents, Anthropic's AI development capabilities are surpassing those of big tech companies. The next-generation language model "Claude 4" released by Anthropic last month is being hailed as the most powerful among commercial AI models. Notably, the higher version "Opus 4" scored 72.5% on the SWE-Bench measuring coding capabilities, outperforming ChatGPT's latest version GPT-4o (69.1%). Researchers explain that Claude possesses the abilities of complex long-context reasoning, external tool integration, and self-memory management, making it a "truly agent-like model" with a technical credibility advantage. The popularity of "Claude," which developers directly use, is believed to influence talent influx. SignalFire's report states, "Engineers are naturally drawn to companies that create products they frequently use and admire."
The fact that Anthropic is a startup also adds to its appeal. OpenAI, founded in 2015, has exceeded a corporate valuation of $300 billion (approximately 407 trillion won), while Anthropic's current corporate valuation stands at around $61.5 billion (approximately 84 trillion won). This perception that there is an opportunity for young talents to secure an initial stake in a potentially larger growth organization is strong. SignalFire noted that "Anthropic possesses an organizational culture that encourages researchers' autonomy and intellectual discussions, acting as a stronger incentive than the salaries and brand power of big tech," adding that "practical work environments, such as flexible working conditions and fast promotion paths, are also advantages."