Google DeepMind researchers have published a comment in Nature arguing that the rapid deployment of autonomous AI agents demands new ethical frameworks that prioritise human well-being and prevailing social norms. The authors say current guardrails, designed mainly for large-language models operating in controlled settings, are insufficient once systems acquire the capacity to take real-world actions. The call comes amid a broader reassessment of how society should frame and govern artificial intelligence. In a separate blog post, Oxford Internet Institute Visiting Research Fellow Anoush Margaryan contends that AI should be treated primarily as a tool under clear human control rather than as a rival or equal collaborator, warning that anthropomorphising machines risks eroding human responsibility. Start-ups are also experimenting with technical safeguards. Axone’s chief technology officer, Christophe Camel, detailed an on-chain governance protocol that encodes concepts such as delegation, loyalty and sanctions directly into smart contracts, aiming to make rule-breaking either impossible or financially costly. Meanwhile, data-labeling platform PublicAI said it has generated more than $14 million in revenue, emphasising the importance of a 'human layer' to verify AI outputs. Taken together, the publications underscore a convergence of academic, corporate and entrepreneurial efforts to embed accountability into increasingly autonomous AI systems before they scale further into finance, healthcare and other high-stakes domains.