Ex-OpenAI, Anthropic Staff Advocate for AI ‘Right to Warn’

Former employees from some of the top companies in artificial intelligence are calling for enhanced whistleblower protections within their organizations. They believe this would enable them to communicate “risk-related concerns” about the development of advanced AI systems to the public more effectively. On June 4, a group of 13 current and former employees from OpenAI, Anthropic, and DeepMind, along with well-respected AI researchers like Yoshua Bengio and Geoffrey Hinton and noted AI scientist Stuart Russell, launched the “Right to Warn AI” petition. The initiative aims to secure commitments from AI companies to allow employees to raise red flags about AI risks both internally and publicly.

William Saunders, a former OpenAI employee and advocate for the petition, emphasized that it’s crucial to create channels for sharing information about potential dangers with independent experts, governments, and the general public. According to Saunders, those most knowledgeable about how advanced AI systems function and the associated risks are often unable to speak freely due to fears of retaliation and overly restrictive confidentiality agreements.

The “Right to Warn AI” proposal outlines four key recommendations for AI developers. First and foremost, it advocates for the removal of non-disparagement clauses specifically related to AI risks. This would ensure employees are not silenced by agreements that prevent them from speaking out about potential dangers. The proposal calls for the establishment of anonymous reporting channels so that concerns about AI risks can be raised safely and effectively. It seeks to protect whistleblowers, ensuring that employees who disclose important risk-related information are not subjected to retaliation.

Saunders described these principles as a proactive step to engage with AI companies to develop safe and beneficial technologies. The goal is to foster an environment where concerns can be openly discussed without fear of punishment.

This petition highlights growing worries about the perceived “deprioritization” of safety measures in AI labs, especially as these organizations work toward creating artificial general intelligence (AGI)—software capable of human-like thinking and self-learning. Daniel Kokotajlo, another former OpenAI employee, mentioned that he left the company because he became disillusioned with their approach to responsible AI development, particularly in the context of AGI. He criticized the industry’s “move fast and break things” mindset, arguing that such an approach is highly inappropriate for technology as powerful and misunderstood as AI.

Former OpenAI board member Helen Toner recently shared similar concerns during a Ted AI podcast on May 28. She disclosed that Sam Altman, OpenAI’s CEO, was reportedly dismissed from the company for allegedly withholding critical information from the board, further highlighting governance issues within the organization.

The creation of the “Right to Warn AI” petition indicates a growing consensus that addressing ethical concerns and ensuring robust safety mechanisms in AI development are paramount. This movement seeks not only to protect employees who wish to speak out but also to inform the public and relevant authorities about potential risks.

As AI continues to evolve rapidly, the call for more transparency and accountability becomes ever more pressing. Current industry practices may not be sufficient to handle the complexities and potential dangers posed by advanced AI systems. Hence, fostering an environment where employees can voice their concerns without fear of retribution is a critical step toward achieving safer AI technologies.

The petition underscores the pressing need for a balanced approach to AI development—one that prioritizes safety and ethical considerations as much as technical innovation. Through better whistleblower protections and more open dialogue, the industry can work toward creating AI systems that are both advanced and safe for widespread use.