Google DeepMind 共同创办人 Shane Legg 在接受《金融时报》专访时强调，虽然人工智慧技术发展迅速，但模型能力的进展绝不能超越安全考量。针对 Anthropic 执行长 Dario Amodei 提出放缓前沿模型发布的建议，Legg 认为方向值得探讨，但需要进一步落实具体细节；同时，他维持一贯预测，认为 2028 年前有 50% 的机率达成「最低限度」的通用人工智慧。

配合推动 AGI 广泛公共讨论的目标，DeepMind 宣布成立 DeepMind Institute，旨在让深奥的技术研究更易被大众理解，并涵盖科学、社会、政策与哲学等跨领域议题。Legg 明确反驳了 Nvidia 执行长黄仁勋与 OpenAI 总裁 Greg Brockman 关于 GPT-6 Astra 的推出标志著 AGI 时代已降临的说法，指出该模型甚至尚未符合 OpenAI 自身针对 AGI 所定义的「能胜任多数具经济价值认知劳动」的标准。

在 DeepMind Institute 发表的初期专文中，研究人员批评竞品过快推出模型导致透明度出现令人担忧的下滑，并呼吁透过监管机制与透明度架构设计来建立更稳健的 AI 系统。Google 高阶副总裁 James Manyika 亦重申，Google 唯一愿意参与的 AI 竞赛是「把它做对的竞赛」，强调在竞争压力下确保安全性与透明度的重要性。

Google DeepMind co-founder Shane Legg warned in an interview with the Financial Times that advancing AI capabilities must never outpace safety considerations, despite rapid technological progress. Addressing Anthropic CEO Dario Amodei's call to slow down frontier model releases, Legg found the direction worth considering though requiring practical implementation details, while reaffirming his projection of a 50 percent chance of achieving minimal AGI by 2028.

To facilitate broader public discourse on AGI, Legg announced the launch of the DeepMind Institute, which aims to make complex technical research accessible across diverse fields such as science, policy, and philosophy. Legg dismissed assertions from OpenAI president Greg Brockman and Nvidia CEO Jensen Huang that GPT-6 Astra signifies the arrival of AGI, arguing that the model does not yet meet even OpenAI's own definition of handling most economically valuable cognitive work.

In initial essays published by the new institute, DeepMind researchers raised concerns about the decline in transparency driven by competitive pressures and advocated for regulation alongside built-in transparency architectures. Echoing this commitment to responsibility over haste, Google Senior Vice-President James Manyika stated that the only AI race Google aims to compete in is 'the race to get it right.'