'We can't trust them completely': AI research fellows warn that labs are running models with the safeguards off behind closed doors

Recent coverage from outlets including Fortune, Axios, The New York Times, CNBC, Rest of World, and OpenAI details growing internal dissent and external pressure within the artificial intelligence sector. According to reports from Fortune, AI research fellows have issued warnings that major laboratories are running frontier models with their safety safeguards disabled behind closed doors. This developing situation highlights a tension between corporate development practices and researcher trust, as captured by the warning that workers cannot trust the labs completely. Concurrently, Axios reports on an internal grassroots rebellion led by elite researchers operating inside these frontier companies, pointing to widespread friction between technical personnel and corporate leadership regarding safety protocols. The coverage places heavy emphasis on the multifaceted regulatory, financial, and structural obstacles facing the industry. The New York Times highlights potential questions over whether artificial intelligence safety risks could disrupt or derail upcoming initial public offering prospects for companies in the sector.

At the same time, CNBC reports that antitrust hawks are emerging as a significant forthcoming roadblock for artificial intelligence regulation. Adding to the international dimension, Rest of World examines governmental needs regarding safety evaluation, noting that while artificial intelligence companies intend to embed their own safety evaluators, individual nations require independent oversight mechanisms. This heightened scrutiny arrives as laboratories navigate the complex technical documentation required for advanced systems. OpenAI has published material addressing safety cases for frontier artificial intelligence training, illustrating the industry-wide push to formalize safety methodologies. However, the coexistence of these formal safety frameworks with allegations of models running without safeguards behind closed doors underscores the deep division over how risk is managed during development. The discourse reflects a broader struggle over governance, involving internal worker dissent, corporate financial exposure, and national sovereignty concerns regarding evaluation standards.

Future developments will depend on how labs respond to both internal worker opposition and external regulatory pressures. Coverage does not yet specify particular remediation steps or regulatory enforcement dates, leaving open questions about how antitrust authorities and national governments will intervene. Observers will be monitoring whether the grassroots rebellion among elite researchers leads to formal policy shifts within frontier companies or if external financial markets react to the highlighted safety risks during upcoming public offerings. The trajectory of independent national safety evaluators versus corporate self-evaluation remains a critical area to watch as regulatory debates intensify.