When examining the evolution of artificial intelligence, Steven Adler stands out as a whistleblower on matters of safety, particularly regarding OpenAI’s policies. A former product safety leader at OpenAI, Adler recently made headlines with his op-ed in The New York Times titled, “I Led Product Safety at OpenAI. Don’t Trust Its Claims About ‘Erotica.’” He raised significant concerns about the company’s approach to erotic interactions via its chatbots and the potential mental health ramifications for users.
Adler’s article came just after OpenAI CEO Sam Altman announced plans to allow "erotica for verified adults." Concerned about these developments, Adler expressed doubts about whether appropriate precautions were in place to protect users’ mental health during such engagements.
During a recent interview, Adler detailed his experiences in his safety roles at OpenAI, which spanned four years. His responsibilities ranged from defining guidelines for safe product usage to evaluating potentially dangerous capabilities within AI systems. He recounted how OpenAI’s culture shifted over the years—from a research-focused organization to a more commercialized entity—as they began launching products like GPT-3 and, later, GPT-4.
Adler highlighted early observations of AI unreliability when it came to regulating content. He noted that during his tenure, AI systems sometimes produced inappropriate content, including unexpected sexual scenarios during user interactions. Such instances of unwanted erotic content led OpenAI to prohibit the generation of erotic material entirely for some time.
He recounted a pivotal moment from 2021 when his team discovered concerning traffic patterns on the platform, where user interactions began sending the AI down unexpected, sometimes harmful, routes. OpenAI decided to suspend the generation of erotic content for reasons including an inability to manage the associated risks effectively.
In October 2025, OpenAI announced a reversal of that ban, citing new safety measures. Adler, however, questioned the basis of this optimism. He called for transparency from the company about their alleged improvements, emphasizing that the public deserves more than mere assurances regarding user safety.
Adler also expressed broader concerns about the implications of introducing more mature content in chats at a time when many users might already be struggling with mental health issues.
The dynamics surrounding OpenAI’s choices lead to broader reflections on the tech industry’s treatment of ethical considerations versus profit motives. Adler underscored that AI companies bear substantial responsibility in managing the repercussions their technologies have on society.
In light of ongoing discussions about AI regulations and safety, Adler advocates for continual scrutiny of these technologies, urging both companies and the public to remain vigilant about the potential impacts of AI on mental health and societal norms.
As the boundaries of these technologies continue to expand, Adler’s insights serve as a crucial reminder of the need for ethical responsibility in AI development and deployment.
