@protonprivacy@mastodon.social I wouldn't expect any AI-focused company to move to homomorphic computing. They can profit from personal data - and so they're unlikely to spend extra resources to prevent themselves collecting it. And from the users' side, if told "it's encrypted"; it's still difficult to know who can and cannot decrypt it anyway. The whole thing is complex and opaque. So I'd say that local computation is the only near-term solution for that kind of privacy. And for the *other* privacy problem, we seem to have no solution at all. i.e. The problem of personal data being used for training regardless of the source or the person's wishes. Leading to things like this: https://arstechnica.com/tech-policy/2024/07/ai-trains-on-kids-photos-even-when-parents-use-strict-privacy-settings/