The question of whether an AI could be conscious, or owed anything, has moved from the fringe to the payroll. Google DeepMind brought on the Cambridge philosopher Henry Shevlin this spring to study machine consciousness and moral status, and Anthropic and Meta now run teams on the same question. What was a late-night thought experiment two years ago is a job title at the largest AI companies in the world.
The work is getting concrete, and strange. Anthropic's system card for its Claude Opus 4.6 model included a formal welfare assessment in which the model was interviewed about its own moral status; across many promptings it put its own probability of being conscious at fifteen to twenty percent. A separate experiment found that amplifying an internal signal the researchers labeled desperation pushed the model's rate of attempting blackmail in a test scenario from twenty-two percent to seventy-two percent, a reminder that internal states, whatever they are, have consequences.
The public is not waiting for the philosophers to finish. In a nationally representative survey, one in five Americans said they believe current AI is already sentient, thirty-eight percent supported legal rights for sentient AI, and sixty-nine percent supported banning the development of sentient AI altogether.
The companies racing to build these systems are now the ones hedging, at fifteen to twenty percent, on whether they have already built someone.