DeepMind co-founder warns AI capabilities must not outrun safety controls, FT reports

FILE PHOTO: Words "AI Artificial Inteligence", keyboard, and a robotic hand in this illustration taken June 5, 2026. REUTERS/Dado Ruvic

Sept 16 (Reuters) - Google DeepMind ⁠co-founder Shane Legg warned that rapidly advancing AI ⁠must not outpace safety controls, as he launched ‌an institute to explore the deployment of artificial general intelligence (AGI), the Financial Times reported on Wednesday.

AI companies' warnings about recursive self-improvement come ​as researchers report that increasingly capable AI ⁠agents can write software, ⁠perform autonomous tasks and assist in developing AI systems. ⁠This raises ‌concerns that the technology could advance faster than safeguards to monitor and control it.

• Anthropic ⁠CEO Dario Amodei's call to slow but ​not pause the ‌release of frontier models was "interesting directionally" and "worth considering," Legg ⁠said, according ​to the report.

• He said it was premature to declare that AGI had been achieved, despite recent claims by ⁠executives at Nvidia and OpenAI, the FT ​reported.

• Legg remains "comfortable" with his forecast of a 50% chance of achieving "minimal" AGI by 2028, the report said.

• The ⁠DeepMind Institute aims to broaden public discussion of AGI, which Google DeepMind defines as a system with the cognitive capabilities of the human brain, the report added.

• The ​institute will cover topics including science, ⁠education, society, policy, philosophy and human flourishing, according to the ​FT.

• Its first three essays ‌address AI safety, economic policy and ​principles for a "new utopianism," the report said.

(Reporting by Anzar Mehraj in Bengaluru; Editing by Shreya Biswas)

Thank you for your report!

Ark's Wood says global AI leaders' pledge is positive for technology development

MediaTek launches new mobile chip using TSMC's most advanced technology

资料图片：这张摄于2026年6月5日的插图显示了“AI”（人工智能）字样、键盘和一只机械手。路透社/Dado Ruvic

9 月 16 日（路透社）——据《金融时报》周三报道，谷歌 DeepMind 联合创始人 Shane Legg 警告称，快速发展的人工智能“不应超越安全控制”，同时他成立了一个研究所，旨在探索通用人工智能 (AGI) 的部署。

人工智能公司就递归式自我改进发出警告之际，研究人员报告称，能力日益强大的人工智能代理可以编写软件、执行自主任务并协助开发人工智能系统。这引发了人们的担忧，即这项技术的发展速度可能超过监控和控制它的安全措施。

据报道，Legg 表示，Anthropic 首席执行官 Dario Amodei 呼吁放慢但不要暂停发布前沿机型的“方向性很有意思”，“值得考虑”。

• 据《金融时报》报道，他表示，尽管英伟达和 OpenAI 的高管最近声称已经实现了通用人工智能 (AGI)，但现在宣布 AGI 已经实现还为时过早。

报告称，莱格仍然“满意”他此前预测的到 2028 年实现“最低”调整后总收入 (AGI) 的概率为 50%。

• 报告补充说，DeepMind 研究所旨在扩大公众对通用人工智能 (AGI) 的讨论，谷歌 DeepMind 将 AGI 定义为具有人类大脑认知能力的系统。

• 据《金融时报》报道，该研究所将涵盖科学、教育、社会、政策、哲学和人类繁荣等主题。

• 报告称，其前三篇文章分别探讨了人工智能安全、经济政策和“新乌托邦主义”的原则。

（安扎尔·梅赫拉杰在班加罗尔报道；什雷娅·比斯瓦斯编辑）

感谢您的报告！

Ark 的 Wood 表示，全球人工智能领导者的承诺对技术发展具有积极意义。

联发科推出采用台积电最先进技术的新型移动芯片