Anthropic, OpenAI and Google DeepMind have each brought on dozens of professional philosophers to shape the guardrails that govern their AI systems. They’re drafting “constitutions” of rules—sets of principles that guide how models respond, flag risky requests and handle conflicts. At Anthropic, a team led by Dr. Anna Powers leans heavily on deontological ethics, drawing on Kant’s ideas to forbid lying, coercion or treating people as mere means to an end. That rigidity makes the models’ behaviour more predictable, which Anthropic argues is vital for rolling out robots in homes, hospitals and schools. Other outfits favour a utilitarian approach, weighing outcomes to seek the greatest overall benefit. DeepMind’s ethics group, for instance, built a scoring system that rates potential AI actions by projected harm reduction and social value. Their framework allows trade-offs: a small privacy breach might be acceptable if it prevents a major security threat. The company says this flexibility helps in complex scenarios—like balancing transparency against user safety when moderating online content. These philosophers are not writing lofty manifestos. They collaborate with engineers during fine-tuning, translating abstract rules into training data and reward signals. At OpenAI, ethicist Dr. Miguel Serrano sits in daily model-alignment meetings. He flags subtle biases, proposes new red-lines—no euphemistic persuasion, no emotional manipulation—and tests how well GPT-5 obeys. His metrics track rule violations per thousand queries, with targets under 0.5 for high-risk categories such as medical or legal advice. Big AI labs say this in-house expertise cuts reliance on outside regulation and helps them respond faster to emerging risks. Critics counter that academic philosophers lack technical chops or that all these constitutions still depend on opaque enforcement. But for now, these hires signal a shift: the next frontier in AI won’t just be bigger models or faster chips, but the ethical codes etched into their cores.