But Claude Code, arguably one of the most famous ones, is not. And recently got some heat about sending meta data that wasn't so obvious. Just as a counter-example.
Maybe possibly with --dangerously-skip-permissions. I've been using auto mode and enjoy how it blocks every tool use that could've allowed something potentially sensitive into context. Burns extra tokens though.
Not auto mode itself (that's just Sonnet running a purely security prompt), but if you're working on a slide deck I can see an upload needed so the working model can "look" at the decks to fulfill your request. If that deck wasn't referenced at all in your prompting requirements... well at least we still have sandboxes.