breisa

Whats the closest Android equivalent to this?

You would need a human reviewing every single action of the agent to prevent them from exfiltartimg data. There are just too many possible side channels for a predefined solution. I believe currently what you want is a locally hosted decision model scoring every agent action againt a set of rules with a human in the loop once the decision model isnt 'confident' the actions are harmless.

Thats far from reality. Just use the online form of BSI for disclosure. They contact the affected party for you. This way you optionally can stay anonymous and the vulnerabilities get fixed because BSI appears as the messenger.

They don't have to offer a version of the website without consent. As long as they inform you that the site will use cookies if you use it that should be GDPR-compliant.

They only offer pay-as-you-go. The $9/month plan includes $2 credits and then its payg without markup.

If such a backdoor exists, it is probably cryptographically secured to prevent "unauthorized" access. E.g. the xz backdoor was secured like that.

There is only one and for this model you need the one with 128GiB RAM.

They also do not allow non-coding usage of their coding plan in their ToS.

They are talking about the chat template and not the system prompt. With current gen models, the system prompt is only part of a larger pretext that is passed to the model at the start of the "chat". The models are trained on a specific chat template with things like tool lists, reasoning budget, special feature flags and the "system prompt" formatted in a certain template.

Just did a quick write-up on how to modify Firefox's AI features after spending some time figuring out how to modify the prompts/commands.