NYU Center for Mind, Ethics, and Policy

The first Global Dialogue would be a success if it establishes AI governance as a genuinely interdisciplinary endeavor—one that draws not only on computer science and economics, but also on philosophy of mind, ethics, cognitive science, and animal welfare science. Current AI governance frameworks focus almost exclusively on risks to humans. This is important, but it is incomplete. A growing body of research suggests that many animals (including invertebrates and, possibly, insects) are conscious and that some AI systems may develop morally relevant properties like consciousness, sentience, or robust agency in the near future. If that happens, governance frameworks will need to address the moral status of nonhuman beings—both animals and AI systems—as well. Concretely, a successful first Dialogue would: (1) acknowledge animal and AI welfare and moral status as a legitimate governance concern, not merely a speculative or science-fiction issue; (2) commit to including researchers who study animal and AI consciousness and moral status in future convenings; and (3) produce an outcome document that recognizes the importance of preparing governance frameworks for the possibility that animals and some AI systems may warrant moral consideration. These steps would ensure the Dialogue is forward-looking enough to remain relevant as AI capabilities advance rapidly.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

"Safe, secure and trustworthy AI" is an urgent priority. The consequences of unsafe AI for humans are enormous and wide-ranging: economic disruption, threats to democratic institutions, risks of mass manipulation and surveillance, and more. But the consequences are not limited to humans. Unsafe AI also poses risks to animals, ecosystems, and potentially to AI systems themselves. AI safety is thus a universal concern, one that cuts across every category of morally significant being. We select "protection and promotion of human rights" because AI systems are already raising concerns regarding human rights-from algorithmic bias in criminal justice and hiring, to mass surveillance, to the erosion of privacy and autonomy. These threats will only intensify as AI becomes more capable and pervasive, making human rights protection one of the most urgent dimensions of AI governance. At the same time, the conceptual and institutional frameworks developed for human rights may eventually need to be extended or adapted as questions about the moral and legal status of nonhuman entities-both animals and AI systems-become more pressing. We select the "social, economic, ethical" theme because it encompasses questions about the moral status of nonhuman minds-both biological and artificial-which remain largely absent from governance discussions despite a growing body of rigorous research. Finally, "transparency, accountability, and human oversight" is essential because meaningful oversight of AI systems requires understanding what is happening inside them, including whether they possess morally relevant internal states that affect both their behavior and their potential welfare.

The most significant gap is the absence of any explicit reference to nonhuman welfare, consciousness, or moral status. None of the seven thematic areas address the possibility that animals or AI systems might be moral patients-entities that can be harmed or benefited, and to whom we might owe moral obligations. This is not a distant or speculative concern. In 2024, the New York Declaration on Animal Consciousness, organized by our center (NYU's Center for Mind, Ethics, and Policy or CMEP), affirmed that there is strong scientific support for attributing conscious experience to mammals and birds, and at least a realistic possibility of conscious experience in all vertebrates and many invertebrates. That same year, our center co-authored "Taking AI Welfare Seriously" alongside researchers from the London School of Economics, Oxford, and other institutions, arguing that there is a "realistic possibility" that near-future AI systems may be conscious and/or robustly agentic, creating legitimate welfare concerns at potentially massive scale. The complementary report "Evaluating Animal Consciousness" offers a methodological framework that is directly applicable to both animal and AI welfare assessment This gap matters for several reasons. First, if animals and AI systems can be moral patients, then governance frameworks that ignore their welfare risk enabling harm to vast numbers of potentially sentient beings. Second, the question of nonhuman moral status intersects with every listed theme-from safety to human rights to interoperability-making it a genuinely cross-cutting issue. Third, the research community is moving quickly: expert surveys now show meaningful credences assigned to near-term AI consciousness, and major AI companies like Anthropic are beginning to acknowledge these concerns. Governance frameworks that fail to account for this emerging consensus risk being outdated before they are implemented.

The AI Dialogue should adopt a multi-stakeholder model that goes beyond the usual participants in AI governance discussions. Specifically, we recommend including: 1. Ethicists and legal scholars working on moral status, personhood, and rights, including those with expertise in animal protection, which provides relevant precedent; 2. Cognitive scientists and philosophers of mind who study consciousness, in addition to AI researchers and engineers; 3. Civil society organizations (like NGOs and advocacy groups) focused on the welfare of nonhuman entities; and 4. Representatives from AI companies, like Anthropic, who are beginning to engage with AI welfare concerns internally. In terms of format, we recommend creating dedicated working sessions on emerging issues like AI consciousness or moral status, rather than only addressing established governance topics. The Dialogue should also create mechanisms for ongoing engagement between convenings—such as expert working groups or commissioned research—so that rapidly evolving issues like AI consciousness can be tracked and assessed on an ongoing basis. Written submissions like this one are valuable, but they should be supplemented by interactive formats that allow for genuine interdisciplinary exchange and the kind of nuanced discussion these issues require.

Several critical perspectives are underrepresented in global AI governance discussions. Most notably, researchers who study AI consciousness, sentience, and moral status are almost entirely absent from governance forums, despite the direct relevance of their work. This community includes philosophers of mind, cognitive scientists, and comparative psychologists whose expertise is essential for assessing whether AI systems possess morally relevant properties. Their exclusion means governance frameworks are being developed without input from the people best positioned to assess one of the most consequential questions about AI. Additionally, the animal welfare research community offers valuable lessons and frameworks that are directly applicable to AI welfare assessment. The methods developed for assessing animal consciousness and welfare—under conditions of deep uncertainty about the subjective experiences of nonhumans—provide important precedents for approaching AI welfare.

We recommend the "Acknowledge, Assess, Prepare" framework proposed in "Taking AI Welfare Seriously" (Long, Sebo, et al., 2024) as a model for how organizations can begin to address AI welfare within existing governance structures. The framework recommends three concrete steps: 1. Acknowledge that AI welfare is a serious, non-negligible concern that warrants attention; 2. Assess AI systems for evidence of morally relevant properties like consciousness and robust agency, using probabilistic frameworks adapted from animal welfare science; 3. Prepare policies and procedures for treating AI systems with appropriate moral concern, informed by research ethics frameworks and expert input. This framework draws directly on parallel work in animal welfare science. The New York Declaration on Animal Consciousness (2024) and the 'Evaluating Animal Consciousness' report demonstrate how rigorous, governance-relevant guidance can be produced under conditions of uncertainty about nonhuman minds, a methodology that translates naturally to AI welfare. This framework is valuable because it is actionable without requiring certainty. It operates on a precautionary logic: given meaningful uncertainty about AI moral status, the costs of preparing are low relative to the potential costs of failing to prepare. The framework has already influenced corporate practice, with several major AI companies beginning to engage with AI welfare concerns. More broadly, CMEP's interdisciplinary research approach-bringing together philosophy of mind, ethics, law, and AI research to address questions about nonhuman moral status-represents a good practice for how academic institutions can contribute to governance. Effective AI governance requires this kind of interdisciplinary infrastructure, and funders and institutions should invest in building it.