Ben explores the tension between corporate values and economic reality, using Anthropic's decision to walk away from a Department of War contract as a lens. When the DoW sought to bypass model safety guardrails, Anthropic declined and accepted the revenue hit rather than compromise its safety standards.
Ben reflects on his own experience watching values erode at a large tech company as revenue pressures mounted, and runs the same thought experiments on Hardshell's work in data-layer security. The same technology that protects sensitive training data could also be used to obscure impermissibly obtained data or mask unauthorized model distillation. Where do you draw the line?
The post also digs into Dario Amodei's essays on AI's potential to reshape health, governance, and geopolitics, and why values matter even more when the systems we build stop behaving deterministically.