OpenAI announced on August 18, 2026, that it is pausing some of its model work due to safety concerns, following a statement from rival Anthropic, which asserted that its safety measures were sufficient to continue operations without a pause. OpenAI's decision comes as both companies prepare for potential initial public offerings (IPOs) and highlights differing approaches to managing AI safety risks.
OpenAI has introduced new safety practices after identifying that its upcoming model, Astra, posed potential cybersecurity risks. CEO Sam Altman indicated that the company would take action if model capabilities exceeded safety and alignment measures. In contrast, Anthropic stated that its safety protocols, outlined in a 186-page report, would prevent the need for a pause on its advanced models.
Historically, Anthropic has been more cautious regarding safety than OpenAI. OpenAI's announcement of slowing the Astra model's release was shared first with Axios, noting that the model may have reached a critical threshold in its preparedness framework. The company is also in the process of updating this framework, which has not been revised since 2023.
Both companies are implementing measures such as releasing models to select partners and slowing the rollout of certain models. However, they are not halting development entirely. The AI industry is currently facing scrutiny over safety following recent cybersecurity incidents involving major AI labs, raising concerns among researchers.
OpenAI is navigating a voluntary federal government review process, the details of which remain undisclosed. Andrew Freedman, co-founder and CEO of AI safety nonprofit Fathom, noted that OpenAI's pause reflects a genuine effort to avoid releasing misaligned models, especially in light of significant staff departures within the company. Former OpenAI board member Helen Toner suggested that the pause could serve as a model for addressing safety concerns in the future.