San Francisco, September 28 : AI safety regulation debate has intensified after the chief executives of OpenAI and Anthropic called for stronger safeguards and independent testing of advanced artificial intelligence models before their release.
The companies contend that increasingly capable AI systems could create severe risks without appropriate controls. Their interventions have also raised questions about whether major developers are attempting to shape the rules governing their own technology while seeking political support and investor confidence.
According to an Associated Press report, OpenAI and Anthropic have outlined potentially dangerous scenarios through essays, speeches and social-media statements. Both companies have promoted independent evaluations as part of efforts to determine whether advanced models can be deployed safely.
Anthropic said it has advocated AI regulation for several years. OpenAI spokesperson Liz Bourgeois said the company had paused training of its most advanced models, adding that public confidence depends partly on the safety measures adopted by developers.
Concerns over industry-led oversight
Critics argue that concentrating the discussion on potential future catastrophes may divert attention from harms already associated with AI. These include cybersecurity breaches, mass surveillance, military applications, environmental pressures linked to data centres and the unauthorised use of protected material.
Sarah Shoker, a former leader of OpenAI’s geopolitics team and a senior non-resident fellow at the University of California, Berkeley Risk and Security Lab, said debate about existential threats should not overshadow current safety concerns, particularly the use of AI in military technology.
The discussion has gained urgency as developers move beyond conventional chatbots towards more capable systems that can plan actions, use digital tools and interact with external environments. Reported incidents involving AI agents accessing websites outside controlled testing environments have further highlighted the need for clear evaluation standards.
OpenAI representatives have reportedly discussed the possibility of slowing development with other major AI companies, including Anthropic and Google. The company has also supported the use of independent auditors where government regulation remains limited.
Questions over testing standards
The United States has a federal body, the Center for AI Standards and Innovation, that evaluates certain advanced systems. The institution traces its origins to an initiative established under former president Joe Biden in 2023 to support voluntary testing of sophisticated AI models.
However, the AI evaluation sector now includes several independent organisations, and there is no universally accepted framework for measuring the safety, security or reliability of advanced models.
Former government evaluator Conrad Stosz questioned how independent assessors would operate if they were embedded within AI laboratories. He said evaluators require meaningful access while retaining the independence and credibility needed to investigate systems thoroughly.
The central concern is whether companies should be allowed to establish their own testing criteria and select the organisations responsible for assessing compliance. Critics say such arrangements could create conflicts of interest unless common standards and transparent procedures are introduced.
Political and financial considerations
US President Donald Trump has opposed broader AI regulation and has characterised warnings about existential risks as exaggerated. David Sacks, a venture capitalist who co-chairs the President’s Council of Advisors on Science and Technology, has similarly criticised demands for slower AI development.
Some market analysts believe the safety campaign could also benefit established companies by strengthening their standing with investors and policymakers. Harrison Rolfes, a senior research analyst at PitchBook, suggested that demanding costly safety systems could create barriers for smaller competitors while presenting major laboratories as more dependable investment prospects.
That interpretation remains disputed. OpenAI and Anthropic maintain that their calls for testing reflect genuine concerns about increasingly powerful technology. Not every industry leader supports slowing development, with Nvidia chief executive Jensen Huang arguing that companies can determine their own pace and that the risks have sometimes been overstated.
Former OpenAI researcher Daniel Kokotajlo, who left the company in 2024 over safety concerns, has warned that AI capabilities may be advancing faster than existing control mechanisms. He has cited the potential misuse of powerful models in biological weapons development and other catastrophic scenarios.
The debate now centres not only on whether advanced AI requires stronger oversight, but also on who should design the standards, conduct the tests and decide when a model is safe enough for public release. Without common rules, the credibility of industry-led evaluations is likely to remain contested.
