Mozilla have taken into consideration doing things locally, such as tab organisation and the likes (one would assume pre-GPT era and with regard to features not utilising LLMs this would’ve been branded as ML functionality) but I’m not fully convinced this still won’t open up potential security issues in the future[for users of AI browsers].
For Tor users this seems even more of an issue as one would expect nation-state actors targeting undesirables would look for any potential weak spot to exploit.
Separately I suppose this brings into light how utterly crazy it seems having AI features in the browser chrome versus limited to the website content process/sandbox. It seems like a privacy and security nightmare and now everyone and their gran are releasing “AI browsers”, even the Firefox-based ones inspired by browsers such as Arc and Dia which seem like absolute privacy nightmares.
Seems like slick branding and marketing gets you a pass today when in the past such egregiousness would receive a load of flack cough Avast “secure” browser cough…
Either way good job to the Tor team, I sympathise with how much extra load this adds to each rebase.