Summary of Key Points
This article discusses the experiences of Joshua Achiam, an AI security researcher at OpenAI, as well as OpenAI’s shift from a focus on safety to commercialization. Eight years ago, during Elon Musk’s resignation speech, Achiam criticized Musk’s AGI plans as unsafe, earning him the nickname “stupid donkey” and a golden donkey statue from a colleague. Later, he led OpenAI’s “Mission Alignment Team,” which aimed to ensure the company stayed true to its goal of benefiting all humanity with AI. However, this team was disbanded after just 16 months, and Achiam was promoted to “Chief Futurist” (a position that seemed more ceremonial than substantive). He eventually left OpenAI, which is now preparing for an IPO, resulting in a significant loss of security professionals.
The Golden Donkey Statue: What’s Become of the Young Man Who Challenged the Billionaire?
In 2018, when Musk resigned from the OpenAI board, 25-year-old Achiam had only been at the company for a few months. He interrupted Musk’s speech in front of all employees, stating that Musk’s AGI plans were unsafe and reckless. Musk angrily called him a “stupid donkey.” A colleague (now a prominent figure in the AI community, Dario Amodei, founder of Anthropic) later made a golden donkey statue for Achiam, inscribed with “Never stop being a stupid donkey for the sake of safety.” Eight years later, this statue was used as evidence in a court case where Musk sued someone named Ultraman, but it made Achiam known worldwide as the security researcher who dared to challenge the world’s richest man. Now, Achiam has also left OpenAI.
Where Has OpenAI’s Security Team Gone?
In September 2024, OpenAI established the “Mission Alignment Team” with Achiam as its leader, tasked with ensuring the company’s actions aligned with its goal of benefiting humanity through AI. However, this small team of six to seven people was disbanded quietly in February 2026, and its members were distributed across other departments. Even more troubling is that OpenAI’s other security efforts have also been weakened: Jan Leike, who worked on ensuring AI obeyed human commands, joined Anthropic; Steven Adler, who studied the dangers of AI, and policy expert Miles Brundage left to found an NGO advocating for strict safety standards; Andrea Vallone, who handled ChatGPT’s emotional response capabilities, also moved to Anthropic. OpenAI’s spokesperson claims that “security responsibilities are embraced by the entire company,” but it’s clear that the dedicated security professionals are no longer there, and the priority for safety has declined.
Achiam’s View on AI Safety: Not a Believer in Doomsday, But Also Not a “Blind Optimist”
Achiam does not belong to the extreme camp that predicts AI will eradicate humanity within ten years. He criticized Eliezer Yudkowsky, a prominent doomsday theorist in the AI community, saying, “Your extreme views are useless and your predictions are all wrong.” He believes the probability of human extinction within ten years is only one in a million and calls those who claim a 5%-10% chance “unaware.” Yet he supports AI regulation and safety measures, though he dislikes the moral high ground often occupied by security advocates. Interestingly, Achiam has sold at least $10 million worth of OpenAI stocks and still holds tens of millions more—indicating that he is not an idealist detached from reality; rather, he is a pragmatic individual who values both AI safety and the company’s success.
OpenAI Has Changed: From “Safety First” to Preparing for an IPO
OpenAI was originally a non-profit laboratory focused on developing safe AGI. However, with its plans for an IPO, commercialization has become a higher priority. The reduction of the security team and the loss of many security experts are clear signs of this shift. In his resignation letter, Achiam wrote, “The world now knows the truth; one can still pursue the company’s mission outside of OpenAI.” This “truth” might be that OpenAI is no longer a laboratory dedicated to safety but a for-profit technology company. Many of his former colleagues have either joined rival companies like Anthropic or started their own security-related NGOs, indicating widespread disappointment with OpenAI’s transformation.
The Conflict Between AI Safety and Commercialization: Just Beginning
Achiam’s story highlights a core dilemma in the AI industry: should we accelerate commercialization for profit or slow down to ensure safety? OpenAI has chosen the former, and researchers like Achiam either have to compromise or leave. How future AI companies balance these two priorities will likely determine the relationship between humans and AI. After all, with AGI on the horizon, security is no longer a trivial matter. For now, it seems that the forces of commercialization have gained the upper hand.