AI Security Shifts To Data Control

追加: 2026-05-20

[00:00:00]So, so look, I think when when people were first thinking about AI security, I think there was a um an emphasis around kind of vulnerabilities around the models, right? And being able to um uh understand uh if the models were uh problematic, if they've been updated. But, the reality of today is that most of the models are not your own. The models are commercial models. Uh they're from Open Open AI, they're from Gemini, they're from uh Microsoft uh with Copilot, they're from um obviously Anthropic. Uh and maybe a few other companies uh as you know, people are building different kinds of uh models. And and really kind of for most organizations, the only kind of lever that they have is not so much about kind of testing the the vulnerability around um these models uh because at the end of the day, they they get what they get, right? And they don't get upset as another kind of children's poem uh >> [laughter] >> um that uh that I used with with my children uh many many times. Um and so really the only lever that organizations have is what data they share.