Medical AI models can leak sensitive patient information through privacy attacks, even when average risk metrics look acceptable. This paper shows that while aggregate membership inference attack success often appears close to random guessing, individual patient-level risks can be extremely high— with some patients facing near-perfect identifiability.

The risks are also disparate: they disproportionately affect underrepresented groups (by race, disease prevalence, or insurance status). Larger models tend to increase these individual and group-level privacy vulnerabilities.

As AI systems grow more powerful and autonomous, leading AI companies are increasingly hiring philosophers and ethicists to help navigate complex moral, conceptual, and alignment challenges.

Companies like Anthropic, OpenAI, and Google DeepMind are bringing in philosophers to work on ethical frameworks, model behavior, human-AI relationships, and even speculative questions around machine consciousness and “model welfare.” Anthropic, for example, trained its model Claude using an 84-page “constitution” grounded in ethical and philosophical principles.