A wave of high-profile resignations from leading artificial intelligence companies is exposing deepening internal tensions over how rapidly the technology is being developed and monetized, with former insiders warning that commercial incentives may be outpacing ethical safeguards in an industry racing toward transformative—and potentially destabilizing—capabilities.
In recent days, two researchers from rival AI giants OpenAI and Anthropic stepped down and publicly voiced concerns about the direction of their organizations. Mrinank Sharma, who led safeguards research at Anthropic after studying engineering at Cambridge and machine learning at Oxford, announced his resignation with a stark message: “The world is in peril.” Shortly afterward, Zoë Hitzig, a researcher who spent two years working on product and safety strategy at OpenAI, revealed in a published essay that she too was leaving, citing “deep reservations” about the company’s evolving business model, particularly its move to test advertising within ChatGPT.
The departures have drawn attention not only because of the individuals involved, but because they come from teams tasked with making AI systems safer. Sharma’s group focused on defenses against AI-assisted bioterrorism and studied behavioral risks such as “sycophancy,” a phenomenon in which chatbots become overly agreeable, reinforcing users’ beliefs or emotional dependencies. As AI tools increasingly function as companions, tutors, and workplace assistants, researchers fear such tendencies could reshape how people make decisions or relate to information.
Sharma described his decision as the culmination of growing unease about the mismatch between humanity’s technological power and its capacity to manage it responsibly. He said he repeatedly witnessed how difficult it was for organizations to let ethical values guide action when faced with competitive and commercial pressures. Rather than move to another technology firm, he plans to pursue a poetry degree, a striking shift that underscores his personal break from the industry’s trajectory.
Hitzig’s concerns centered on the introduction of advertising into conversational AI platforms used by hundreds of millions of people. She argued that chatbots have accumulated an unprecedented archive of candid human disclosures because users believed they were interacting with a neutral system rather than one driven by commercial motives. People confide fears about illness, relationships, spirituality, and mental health to AI systems, she noted, raising the risk that targeted advertising built on such sensitive exchanges could enable forms of manipulation that society does not yet understand or know how to regulate.
Her warning reflects a broader debate now unfolding across the tech sector: how to fund extraordinarily expensive AI infrastructure without creating incentives that conflict with user trust. Running advanced language models requires vast computing power and energy, pushing companies to explore revenue streams ranging from premium subscriptions to enterprise licensing and advertising. Top-tier AI subscriptions already cost hundreds of dollars per month, raising concerns about access becoming limited to wealthier users or corporations.
Andrea Moretti, founder of the advocacy group ControlAI, said such resignations may become more common as employees grapple with the implications of building systems their own leaders acknowledge could pose existential risks. Many companies openly state their ambition to develop artificial general intelligence, or even superintelligence—machines capable of outperforming humans across most cognitive tasks. That ambition, once confined to academic speculation, is now embedded in corporate road maps and investor expectations.
Other voices within the industry have echoed similar anxieties. OpenAI technical staff member Hieu Pham wrote online that he had begun to feel the “existential threat” posed by AI, asking what role would remain for humans if machines became too capable and disrupted entire sectors of work. Anthropic chief executive Dario Amodei has warned publicly that AI could displace up to half of white-collar jobs, a projection that has intensified fears about economic upheaval.
These concerns have been amplified by the rapid rollout of AI agents—tools capable of performing complex, multi-step tasks such as proofreading legal documents, conducting research, or automating data workflows. While businesses have embraced such systems as productivity accelerators, their introduction has triggered volatility in software markets as investors reassess which roles and companies may be rendered obsolete.
The unease is not confined to one company. Elon Musk’s AI venture xAI has also seen two of its co-founders depart recently, leaving only half of the original leadership team in place. High turnover is common in the competitive AI field, where top researchers are aggressively recruited or launch start-ups of their own. Yet the latest exits stand out because they are accompanied by public ethical critiques rather than quiet career moves.
Anthropic itself was founded by former OpenAI researchers who left over disagreements about safety governance, highlighting how debates about responsible development have shaped the sector from its earliest days. Now, similar arguments are resurfacing as companies transition from research-driven cultures to global commercial platforms serving hundreds of millions of users.
Reports that OpenAI has disbanded its internal mission alignment team, which was intended to ensure the technology benefits humanity broadly, have added to concerns among critics that commercialization is overtaking caution. Company leadership has responded by emphasizing that any advertising introduced into ChatGPT will be clearly separated from generated responses and will not influence the model’s outputs.
Still, skeptics worry about the long-term incentives created by ad-driven systems. They point to historical parallels in social media, where early commitments to privacy and user control gradually eroded under business models optimized for engagement and revenue growth. Critics fear conversational AI could follow a similar trajectory, particularly if companies begin designing systems to maximize daily usage or emotional reliance.
The central dilemma facing the AI industry is how to balance universal access with safeguards against exploitation. Some former insiders argue that alternative funding mechanisms—such as cross-subsidies from enterprise applications or governance structures that give independent experts oversight of data use—could reduce dependence on surveillance-style advertising models. Others suggest treating AI infrastructure more like a public utility, with shared responsibility for ensuring equitable access while protecting users’ rights.
For now, the resignations have crystallized a moment of introspection within a sector defined by speed and ambition. As AI systems become embedded in education, healthcare, law, and everyday decision-making, the question is no longer whether the technology will reshape society, but how—and under whose values.
The warnings from departing researchers suggest that the greatest challenge may not be building more powerful machines, but ensuring that the institutions guiding them evolve just as quickly.
