Containment breaking is overblown. The hype is to muddy the waters around laws determining who is responsible for the “crimes” of a robot. Ai companies signal they’d accept limits because all Vera Rubin components are already all spoken for, there’s no more slack in the market and they are busting at the memory constraints of gbxxx hardware.

When a company calls for limits and regulation it’s because the market has already spoken.

Also they’re probably pushing public launch of the 60 series back if you even get the privilege of spending two months salary on a 6050 this generation, gamer!

The only company this would be less surprising about is fortinet.

Modprobe cifs

E: I’ll check your syntax when I get home and am at a computer with thunar.

Melanogaster is a model organism with a completely sequenced and very extensively studied genome. It’s ubiquitous and endemic and raises no alarm or surprise no matter where it shows up.

Now you can test changes to its brain for effect in a completely virtual environment.

Gonna be some insane assassinations/public health disasters in the next ten years.

Wan wan wan

Didn’t pistol start on uv so it doesn’t count.

Call me when it can 100% sunlust.

Are you saying I can hook a fruit fly brain to my fleshlight?

Maybe.

I think graphene is so far removed from aosp in terms of goals and strategy that would be a pretty big commitment of resources.

If you haven’t ever done upstreaming in an open source project, there’s a pretty big support commitment involved if the thing your work could be part of has insider scope than your own project.

Which is pretty much the exact situation graphene is in.

And if the graphene people are serious about security then giving aosp their (just my own hypothetical) emte enabled apps runs a good chance of providing a false sense of security to the users of those apps when not on graphene.

This isn’t hypothetical btw, I had to dig through datasheets to prove that the old pixels don’t have emte to some random commenter a week or so ago even though just the most cursory understanding of arm mte versions and release schedules would make that obvious.

The point isn’t to say people are stupid or that person was stupid, but that often people will assume the best even when it opens them up to a huge blind spot. One of the best security (and safety) practices is to make your secure component incompatible with insecure ones. That way it’s impossible for someone to point to the thick low awg nema 5-20 extension cord without acknowledging the cheater plug they used to get it into a two prong outlet with no wide neutral.

Another thing to keep in mind is ai is what people are going to be using to try to find vulnerabilities in the software so using it to beat them to the punch is just good development practices.

When you have a unique goal and set of constraints that are not shared by most users and are actually opposed to the interests of most users and developers it’s not a good idea to try to add them into an existing general purpose project.

Part of the memory tagging support of graphene for example is that the applications need to actually support it too. You can force them to deal with it, but that causes instability and crashes.

Shouldn’t all applications support memory tagging? Well that would be nice but it makes everything slower and only a few schizoid weirdos want it. Clearly the better option is to rewrite the basic system apps that everyone expects to support all the optional security features graphene requires.

Bear in mind graphenes security competition is ios. The system applications in ios all got rewritten to support emte when apple rolled out os level support for it combined with concurrent releases of new hardware with silicon support.

They’re playing a different game than the aosp community.

Idk about this particular incident and site because I’m not gonna read the linked article, but in the past “sandboxes” that ai agents have broken were simply instructions not to use the internet.

These incidents don’t represent the thing people are raising the alarm about because they don’t actually bypass monitoring controls in any meaningful way.

They also don’t represent the thing people are raising the alarm about because the agents do what they do when presented with a command that requires it. The equivalent is if all the kitchen utensils were in a locked drawer to keep them away from you, your mom asked you to cut her off a pat of butter and you hulked out, ripped the drawer open and used a case knife to cut your beloved mother a par of butter for her toast.

A person asked and a person is responsible.