From Winter v. GP Putnam, here are cites to some of the existing cases at that time; there are further ones but the reasoning of the 9th here makes there relevance explicit so wanted to provide that.

That passage suggests that software is potentially a "product" to the extent that it doesn't give the result intended by the designer. Since OpenAI does try to prevent harmful output — and anyway wouldn't want to admit that it intended this output — that puts them in a bad position here. No?

No it suggests that if its a specialized technical tool, that's the case