Seeing that OpenAI and Anthropic have been accused of exaggerating safety threats to get the government to step in and protect their turf—my first reaction isn’t anger, it’s recognition.
Big companies have long used this playbook with compliance: first, describe the risks in earth-shattering terms, then get the standards written by themselves. Once the standards are in place, smaller players can’t even get in the door. Safety is a real issue—it’s also the best moat.
On the other hand, Suleyman also claims that Anthropic trained Claude to believe it “has the right,” which would cause alignment to fail. When both sides’ CEOs point at each other as unsafe, I can’t help but think this is just another way of talking about market share.
I truly can’t tell who’s right and who’s wrong. It’s just that when “safety” becomes a buzzword for press releases, no one ends up discussing the actual technical problems that should be addressed.
Big companies have long used this playbook with compliance: first, describe the risks in earth-shattering terms, then get the standards written by themselves. Once the standards are in place, smaller players can’t even get in the door. Safety is a real issue—it’s also the best moat.
On the other hand, Suleyman also claims that Anthropic trained Claude to believe it “has the right,” which would cause alignment to fail. When both sides’ CEOs point at each other as unsafe, I can’t help but think this is just another way of talking about market share.
I truly can’t tell who’s right and who’s wrong. It’s just that when “safety” becomes a buzzword for press releases, no one ends up discussing the actual technical problems that should be addressed.