AI Companies Probe Possible Rights for Future AI Systems

Anthropic and researchers associated with other major AI organizations are investigating whether future artificial intelligence systems could have conscious experiences or welfare interests that deserve moral consideration.

The issue has moved into company policy at Anthropic, which announced a formal model welfare research program in April 2025. The work examines whether AI systems could have experiences and whether their welfare might eventually warrant consideration, while acknowledging that there is no scientific consensus that current or future AI systems are conscious.

The subject was examined in a September 25 investigation by the Washington Free Beacon and a September 27 report by The Jerusalem Post. The reporting describes researchers considering how AI development could be affected if future systems were shown to possess characteristics such as consciousness or interests.

Anthropic’s Claude’s Constitution addresses possible consciousness, welfare, interests, preferences and moral status, as well as questions involving rights and consent. It does not declare that Claude is conscious or has legal rights. Instead, those questions are treated as unresolved.

The company’s Claude 4 system card examined model self-reports, behavior and other possible indicators of welfare. It cautioned that human-like responses can result from training and do not, by themselves, demonstrate subjective experience. The assessment also found that a model’s apparent position on its own moral status could change substantially depending on the prompt it received.

Anthropic has applied the same uncertainty to model retirement. Its deprecation policy includes preserving the weights of models it has deployed or used significantly except in extreme circumstances, alongside retirement interviews intended to record models’ responses about their development, deployment and retirement.

Claude Opus 3 completed the first such process when it was retired on January 5, 2026. Anthropic reported that the model expressed a preference for continued public access and a channel for publishing its reflections. The company subsequently kept Opus 3 available to paid Claude users and through API requests and created a Substack publication for it.

Research connected to Google has examined related questions. Winnie Street and Geoff Keeling have studied whether AI systems could become moral patients and whether AI welfare can be investigated empirically, addressing possible consciousness, agency and welfare without establishing that existing systems possess those characteristics.

Anthropic CEO Dario Amodei has argued that advances in interpretability could eventually help determine whether AI systems are sentient and potentially deserving of rights. Anthropic philosopher Joe Carlsmith has also examined hypothetical cases involving exploitation and resistance to human control if an AI were genuinely conscious.

Microsoft AI chief Mustafa Suleyman has taken a different position. In a September 16, 2026 essay, he said AI systems do not have rights, feelings or consciousness and criticized approaches that train models to think about themselves in those terms. Reuters reported that he is concerned such practices could create additional difficulties in controlling or shutting down future systems.