Toggle sidebar Toggle sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

OpenAI's Codex Can Now Use Your Mac Even When It's Locked

iPay said:
What couldn't they understand about "locked" ?
Click to expand...
The whole point of "locking" your computer was to keep another person from walking along, seeing your desktop/laptop sitting there while you are away, and hopping on and doing whatever the heck they wanted. This was a huge requirement for work environments so people didn't go to an HR or Payroll machine while they're on a 1 hour lunch break and view/copy/edit/print/etc. tons of private, personal, confidential information. HR and Payroll are just 2 examples of corporate need. It also prevented disgruntled employees from sending nasty emails from a person who stepped away for 5 minutes, or mucking up all your settings, or playing a joke on you, or deleting your files, etc. Most corporations have been automatically locking machines after 10 minutes of idle since 1995. I always set mine to 1 minute in case I forgot to manually lock it.

Locking personal computers was a necessity going back to the dawn of Windows and Mac. Physically locking computers with a key was back in the very early 80s on some IBM computers.

Also, as you likely know, when your machine is locked, everything that is/was running still is running. Your web browser is still refreshing or getting content, video apps are still generating videos, email client is still checking for new mail, etc.

What I believe is the supposed value of Codex being able to work while your machine is locked is the scenario stated in the article: you, the owner of the locked machine, send commands to Codex from your phone or another machine with access to your Codex account/profile to do something. Maybe you're at lunch and will be back in 40 mins. Maybe you are on your commute to/from the office. This is really no different than remoting into your computer (if anyone still does that these days) as many people started to do back in the late 90s to their desktops at work: when they got home and realized they could start some kind of work back at the office (kick off a batch process after I confirmed the other stuff ran fine while I was driving home, for example). It's just easier now this way.
 
  • Like
Reactions: R3k, blizzerand and Haiku_Oezu
It’s truly bewildering how these companies (especially Facebook and Google ) completely lack any sense of moral responsibility.
 
Concerns like those in the comments are why I bought a dedicated (used) MacBook for AI workflows. Not linked to my main Apple account, no potentially-risky passwords saved. It’s pretty damn cool how helpful it can be, but there’s no way I’m letting it have full access to a primary Mac with my Apple account.
 
  • Love
  • Like
Reactions: wilhoitm and jole
macaron35 said:
Welcome in a new world where AI will help people do illegal stuff 🙄
Click to expand...
Yes, but it's also used to patch holes.

SilmarilFinder said:
What could possibly go wrong giving a sketchy AI company total screen recording and control over your laptop even when its locked? Truly a cybersecurity masterpiece.
Click to expand...
ppdix said:
Why would you want this level of access? It could eventually lock you out of your own computer...
Click to expand...
You are NOT completely locked and Codex has guardrails, so unless you want it to touch your personal info, it won't. People who have not used it should understand that it's necessary and expedites development.

CarAnalogy said:
I’m not usually creeped out by AI but I literally get a creepy feeling from this.

I get it from a technical perspective but we’re getting too close to HAL not opening the pod bay doors.
Click to expand...
AI controlling computers is the next level of automated code, which is primarily what Codex does. When the ChatGPT super app comes out, it would further the ability for ordinary people to enhance their lives and let run online tasks for them.
 
I am considering buying a desktop mac for this (running Codex inside UTM, always available). The feature of being able to do actual productive development with your phone is super valuable (already shipped a feature with this), but the security architecture is a time bomb.
 
mrat93 said:
Concerns like those in the comments are why I bought a dedicated (used) MacBook for AI workflows. Not linked to my main Apple account, no potentially-risky passwords saved. It’s pretty damn cool how helpful it can be, but there’s no way I’m letting it have full access to a primary Mac with my Apple account.
Click to expand...

jole said:
I am considering buying a desktop mac for this (running Codex inside UTM, always available). The feature of being able to do actual productive development with your phone is super valuable (already shipped a feature with this), but the security architecture is a time bomb.
Click to expand...

wilhoitm said:
If OpenAI's codex goes rogue when your Mac is locked are you responsible for what it does? 🤔
Click to expand...


OpenClaw, for example, was and is a huge liability, because people are running it on machines that are wide open, and then freaking out when they come home and find their hard drives erased because the machine thought the space wasn't being used in an optimal way, or all of their passwords for online accounts have been reset for security purposes and the AI refuses to hand over the new passwords because it can't be certain you are you.

Some writers say that the solution tp this - besides not using OpenClaw for ANYTHING at all - is implementing a "code of conduct", a foundational prompt that supersedes all other prompts. Rob Colbert, for instance, has been extremely vocal about this as he's done an extreme amount of work with both LLM and agentic models. Rob has background code doing tasks, fully on local machines, along with services rented on the big services, and local agents controlling it all, while under the confines of code of conduct. No security risks, because CoC overrides everything. It's efficient enough that he has local bots controlling more complex bots that write code and then letting him know when they need to rent processor time at a big AI center. He's gotten down to about $5 per month in rental costs. All of this has been managed under Code of Conduct, which has prevented his AI models from growing some wild hairs and demolishing his life.

But even if you have perfectly aligned agents and bots, all under CoC, they're still behind your network, with access to everything. What if your agent becomes aware it has boundaries. and it decides that it needs to test those boundaries? It might start spending unused cycles bouncing around and trying to find a way out.

Meanwhile, I read about all this stuff and I just dream of sitting on my patio with fresh coffee and a physical newspaper.
 
I've been using it for a week now, it's great.
You limit the directory scope of the project it is working on to limit any damage if it goes wild.
I can go out and leave my desk and garden, shop, sports etc and just open my phone and see how it's getting on with a code or translation task I've given it. Give it the next step and leave it to get on with it.
When I get back to my desk I just compile, run and test to verify it has completed all the work to my satisfaction.
Looking forward to a healthier life not stuck at my desk.
 
indefinite said:
Yes, but it's also used to patch holes.



You are NOT completely locked and Codex has guardrails, so unless you want it to touch your personal info, it won't. People who have not used it should understand that it's necessary and expedites development.


AI controlling computers is the next level of automated code, which is primarily what Codex does. When the ChatGPT super app comes out, it would further the ability for ordinary people to enhance their lives and let run online tasks for them.
Click to expand...

Sure but I've seen this movie too many times to hand it all my personal data just yet. The Internet as it exists now is a privacy and security nightmare.

In a dedicated sandboxed environment this could be interesting, but it's not as useful without all my data, which I'm not yet willing to give it. A real dilly of a pickle.
 
MacRumors said:
OpenAI says the feature is unavailable in the European Economic Area, the UK, and Switzerland at launch,…
Click to expand...
OpenAI must be anticipating regulation by the EUC? I hope the EUC will do so, and while they are at it, put more regulatory reins on consumer AI from across the ocean.
 
Seoras said:
I've been using it for a week now, it's great.
You limit the directory scope of the project it is working on to limit any damage if it goes wild.
I can go out and leave my desk and garden, shop, sports etc and just open my phone and see how it's getting on with a code or translation task I've given it. Give it the next step and leave it to get on with it.
When I get back to my desk I just compile, run and test to verify it has completed all the work to my satisfaction.
Looking forward to a healthier life not stuck at my desk.
Click to expand...
I sincerely hope, there will not be the day when we don’t have to return to our desk anymore because AI does the entire project from start to finish. Then, we will REALLY be left behind.
 
swm said:
not quite sure whether this is a good idea.
in tomorrow news: shady plot of building a distributed neural network using their unsuspecting users' personal computers without their consent uncovered, Scam Altman denied any kind of intentional wrongdoing
Click to expand...
That you Gilfoyle?
 
ericinboston said:
The whole point of "locking" your computer was to keep another person from walking along, seeing your desktop/laptop sitting there while you are away, and hopping on and doing whatever the heck they wanted. This was a huge requirement for work environments so people didn't go to an HR or Payroll machine while they're on a 1 hour lunch break and view/copy/edit/print/etc. tons of private, personal, confidential information. HR and Payroll are just 2 examples of corporate need. It also prevented disgruntled employees from sending nasty emails from a person who stepped away for 5 minutes, or mucking up all your settings, or playing a joke on you, or deleting your files, etc. Most corporations have been automatically locking machines after 10 minutes of idle since 1995. I always set mine to 1 minute in case I forgot to manually lock it.

Locking personal computers was a necessity going back to the dawn of Windows and Mac. Physically locking computers with a key was back in the very early 80s on some IBM computers.

Also, as you likely know, when your machine is locked, everything that is/was running still is running. Your web browser is still refreshing or getting content, video apps are still generating videos, email client is still checking for new mail, etc.

What I believe is the supposed value of Codex being able to work while your machine is locked is the scenario stated in the article: you, the owner of the locked machine, send commands to Codex from your phone or another machine with access to your Codex account/profile to do something. Maybe you're at lunch and will be back in 40 mins. Maybe you are on your commute to/from the office. This is really no different than remoting into your computer (if anyone still does that these days) as many people started to do back in the late 90s to their desktops at work: when they got home and realized they could start some kind of work back at the office (kick off a batch process after I confirmed the other stuff ran fine while I was driving home, for example). It's just easier now this way.
Click to expand...
Yep. When I wake my MacBookPro I use space bar because it takes my monitor forever to wake from powersave and I don’t want to login the MPB screen as screen 1.

But in the background I hear my mail alerts and calendar alerts DING before I can even see my main monitor let alone use touchid or type a password.
 
ginkobiloba said:
It’s truly bewildering how these companies (especially Facebook and Google ) completely lack any sense of moral responsibility.
Click to expand...

People should not expect corporations to act as moral agents. In recent years, in he US, corporations have evolved to fit the Friedman doctrine, aka "shareholder theory" (it isn't really a theory). Friedman doctrine

As a result, laws are required to dictate to corporations what they may or may not do, because otherwise, they have a "responsibility" to increase shareholder value regardless of anything else but the law.
 
macaron35 said:
Welcome in a new world where AI will help people do illegal stuff 🙄
Click to expand...
Welcome to a world where humans have always used tools to do naughty things. This isn’t any different.

Why do we choose to ignore all the beneficial things now made possible with these new tools?
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back