Anthropic, the artificial intelligence startup known for its Claude models and focus on AI safety, is emerging as one of the most desirable workplaces in the industry.
While some competitors are struggling to retain staff — only three of the 14 authors behind a 2023 paper introducing the Llama model still work at Meta — CEO Dario Amodei’s company is expanding its engineering ranks faster than many of the sector’s biggest names, all without offering compensation packages worth hundreds of millions of dollars and possibly even more than a billion dollars.
Retention advantage amid fierce competition
According to research published in May 2025 from SignalFire, 80% of Anthropic employees hired between 2021 and early 2023 remain with the company two years later. This marks the highest retention rate of all the major AI players: Google DeepMind is at 78%, OpenAI is at 67%, and Meta is at 64%.
Engineers are also eight times more likely to leave OpenAI for Anthropic than the reverse, and nearly 11 times more likely to leave Google DeepMind. “Some of that’s expected — Anthropic is the hot new startup, while DeepMind’s larger, tenured team is ripe for movement,” the SignalFire researchers said. “But the scale of the shift is striking.”
Anthropic’s stats are particularly notable considering Amodei is not willing to match the massive salaries the likes of Mark Zuckerberg are offering to lure the small talent pool of AI engineers over to his team. Amodei said on the Big Technology Podcast last week that doing so would be compromising the company’s “compensation principles, our principles of fairness.”
“If Mark Zuckerberg throws a dart at a dart board and hits your name, that doesn’t mean that you should be paid ten times more than the guy next to you who’s just as skilled, who’s just as talented,” he added.
Anthropic has reportedly lost some engineers to Meta
Nevertheless, at least two engineers — Joel Pobar and Anton Bakhtin — reportedly left Anthropic for Meta’s Superintelligence Labs, though neither has updated their public social media profiles to reflect the move. While Meta can draw from its vast social media revenue to fund new hires and massive data centres, its AI reputation has been clouded by underperforming models, explicit chatbots, lawsuits with copyright holders, and a number of privacy violations.
Amodei said on the podcast that several Anthropic employees turned down offers and “wouldn’t even talk to Mark Zuckerberg.” OpenAI’s CEO Sam Altman also conceded that a “few great people” had joined Meta, but also that “none of (OpenAI’s) best people” accepted offers.
Company mission as a factor for AI researchers’ jobs considerations
Meta is not the only tech giant splashing the cash. Microsoft has reportedly poached around two dozen staff from Google’s DeepMind AI arm to work on AI products like Copilot and Bing AI. It reportedly offers top engineers annual salaries exceeding $400,000 in competitive recruiting scenarios, with on-hire stock packages valued nearly $2 million.
When salary offers from multiple companies reach similar levels, mission and values often become deciding factors for younger AI researchers. Many entered their niche fields long before they were profitable, driven by a belief in AI’s potential to revolutionise industries and improve the world.
Amodei believes this is why Anthropic has managed to retain staff without joining the one-upmanship of the salary wars. “I think that what they are doing is trying to buy something that cannot be bought, and that is alignment with the mission,” he said on the Big Technology Podcast.
Anthropic’s mission has always been safety; the company was formed by former OpenAI engineers in 2021 who were concerned about its increasingly commercial direction and felt that safeguards were falling by the wayside. Anthropic has published a lot of AI safety research, and Amodei has been outspoken about the technology’s risks, leading to him being labelled a “doomer.”
Impressive AI models can also sway AI talent into jumping ship
Company culture is not the only consideration for AI researchers; the capabilities of a company’s AI models can also influence career moves. Both Google DeepMind and OpenAI have claimed that their AI models have reached gold-medal level performance on the International Mathematical Olympiad. Amodei said on the podcast that each of the Claude models has “got substantially better at coding than the last.” Anthropic released Claude Opus 4.1 this week.
OpenAI launched GPT-5 on Thursday with the kind of hype usually reserved for a new iPhone, and Altman said it can “write an entire computer program from scratch.” The company also unveiled two open-weight models — its first since 2019 — in a move that could signal a shift toward wider accessibility.
The question remains: In the competition for top AI talent, will it be mission, money, or model performance that could signal a shift toward wider accessibility?
Apple is reportedly joining the AI chatbot race by developing its own version of ChatGPT.
