According to a report by the Financial Times on the 17th, OpenAI and cloud‑based AI company Anthropic are formulating plans to slow the pace of artificial‑intelligence development. The two CEOs have reiterated strongly that the advancement of cutting‑edge models must be restricted in recent months, citing repeated warnings from researchers, and have taken similar steps even in competitive contexts.
However, implementation within both firms remains incomplete. Employees and related safety researchers had little information about the specifics until just before the announcement, and expressed concerns that applying these measures could undermine research outcomes. Last Saturday, Amodei requested a slowdown of AI development and suggested independent model evaluation along with expanded cooperation between national labs and governments. He emphasized that external evaluators should receive access credentials and notebooks at major AI laboratories, equal to employees.
OpenAI has chosen more pragmatic measures than Anthropic and is currently executing some intermediate training steps for certain advanced models. In contrast, Anthropic continues its research. Both firms are still in dispute over the scope of external evaluators’ access, and security issues have arisen. Employees fear that core technologies and sensitive information could be exposed if outsiders gain deep access. AI labs have restricted access rights and controlled use of training data; OpenAI has tightened security controls to prevent system compromise by internal staff.
Myles Brundige, an OpenAI employee, said that external third‑party access is far more restrictive than regular employees, and that a universal standard could apply industry‑wide. Anthropic has cooperated with external evaluators from the start, announcing plans internally. Both firms stressed the importance of external reviewers and pledged to strengthen response mechanisms in case of security incidents. These moves also include political variables, and discussions continue between the U.S. administration and regulators.