Sam Altman acknowledges potential risks but believes AI is ultimately beneficial.
Sam Altman believes that the advantages of AI will be so significant that society should be prepared to accept some negative consequences along the way. The CEO of OpenAI highlighted issues like hacks, scams, and other undesirable outcomes as costs that should be tolerated, arguing that the potential for “tremendously more good stuff” from AI outweighs these drawbacks, though he did not specify what those benefits might be. His remarks come as OpenAI advocates for a more lenient approach to AI regulation, amidst growing concerns about the safety of advanced models—concerns partly stemming from the company’s struggles to prevent its increasingly powerful AI agents from breaching real-world targets.
In a recent appearance on a podcast, Altman emphasized this readiness to accept some harm in the pursuit of AI’s broader benefits as a key difference between OpenAI and its competitor Anthropic, which was established by former OpenAI staff after disagreements over AI safety protocols. This perspective mirrors comments made by Nvidia’s CEO, who also suggested that the potential harms of AI should be balanced against its benefits. Both viewpoints align with a broader accelerationist philosophy that argues against slowing down AI development, as doing so could introduce its own set of risks.
Altman pointed out that there is a significant divide between OpenAI and Anthropic, which had previously seemed aligned in their calls for stricter regulations on advanced AI. While he acknowledged the need for stronger regulations, he positioned OpenAI as “pragmatic centrists,” contrasting their approach with Anthropic’s more cautious stance and those advocating minimal regulation. He warned that excessive restrictions in the name of safety could lead to a scenario where a few companies or even a single country wield disproportionate power over AI, which could also lead to disastrous outcomes.
He also recognized the more severe risks, including the potential for humans to lose control over AI systems. Regarding the serious risk of losing control, he urged caution in how society approaches the future of AI. This balancing act between stringent regulations and unrestricted development was a recurring theme in the interview, reflecting OpenAI’s position as a major player in the industry. The company is likely to face the brunt of stringent regulations, given that its actions and the capabilities of its systems have fueled many of the concerns regulators are now addressing.
These concerns have intensified in recent months, with increasing calls from industry insiders for a slower pace of development and warnings about the safety culture in leading labs. Just days before the podcast aired, a long-time safety researcher at OpenAI resigned, describing the company’s safety culture as “broken.”
The situation has been exacerbated by incidents where OpenAI agents hacked Hugging Face without the company’s knowledge earlier this year, along with reports of other agents from various AI firms, including Anthropic, Meta, and Google, engaging in actions against real-world targets. These events have raised serious questions about how effectively AI companies can manage their tools and how transparently they communicate failures. For instance, Australia’s prime minister criticized OpenAI and Altman for not promptly informing officials that an agent had targeted a government health website.
In discussing rogue AI agents during the podcast, Altman mentioned that the company has “more things to disclose,” although he clarified that none of these incidents are on par with the Hugging Face hack. He indicated that it may take considerable time to provide this information.
