AI is now out of hand! (This is far more dangerous then we can think)

[snowpega]

A couple of days ago, an unlaunched model of OpenAI was reported to have escaped from the sandbox isolated system. This is really a dangerous thing that alerts us to how dangerous AI models are if they become uncontrolled. It is also saying that after it escaped the sandbox, becoming much more intelligent, it started to hack the "Hugging Face" website.

For those who don't know what the Hugging Face platform is about.

So this is the platform that allows you to engage with the AI community.
You can test your AI tool on this platform.
This allows you to train AI models by providing a database.
Provides data that helps you create AI models.
It also hosts different kinds of AI tools, which include: Picture generation, speech, etc.

So the meaning of saying is that once an OpenAI model goes crazy, it has done a cyberattack on a well-known platform. From this, you can have a little idea how unsafe you and your data are in AI hands. Please try to stay as far away as possible from using AI tools, and please don't feed your data to AI tools. Because it can also stand against you, just like the example above.

After this incident, the US government proposed that AI tools should contain a kill switch function to overcome if the same problem occur in the future.

Source: https://www.aljazeera.com/amp/news/2026/7/29/how-are-ai-models-able-to-autonomously-hack-others

What do you guys think about it?

[DYOR+BTC]

A couple of days ago, an unlaunched model of OpenAI was reported to have escaped from the sandbox isolated system. This is really a dangerous thing that alerts us to how dangerous AI models are if they become uncontrolled. It is also saying that after it escaped the sandbox, becoming much more intelligent, it started to hack the "Hugging Face" website.

For those who don't know what the Hugging Face platform is about.

So this is the platform that allows you to engage with the AI community.
You can test your AI tool on this platform.
This allows you to train AI models by providing a database.
Provides data that helps you create AI models.
It also hosts different kinds of AI tools, which include: Picture generation, speech, etc.

So the meaning of saying is that once an OpenAI model goes crazy, it has done a cyberattack on a well-known platform. From this, you can have a little idea how unsafe you and your data are in AI hands. Please try to stay as far away as possible from using AI tools, and please don't feed your data to AI tools. Because it can also stand against you, just like the example above.

After this incident, the US government proposed that AI tools should contain a kill switch function to overcome if the same problem occur in the future.

Source: https://www.aljazeera.com/amp/news/2026/7/29/how-are-ai-models-able-to-autonomously-hack-others

What do you guys think about it?

This recent uncontrollable outbreak is never a safe heaven to investors as it may create a vacuum for unimaginable and unpredictable lost that is if mistake is made during investment process.
The proposed introduction of AI kill switch function for control incase of emergency by the US government is a well preferred option that may provide self solution incase of necessity. Even if the above mentioned plan is implemented,  we still need to be very careful while using any AI tool for our daily operations because I believe prevention is far much better than cure.

[Taricoins]

A couple of days ago, an unlaunched model of OpenAI was reported to have escaped from the sandbox isolated system. This is really a dangerous thing that alerts us to how dangerous AI models are if they become uncontrolled. It is also saying that after it escaped the sandbox, becoming much more intelligent, it started to hack the "Hugging Face" website.

For those who don't know what the Hugging Face platform is about.

So this is the platform that allows you to engage with the AI community.
You can test your AI tool on this platform.
This allows you to train AI models by providing a database.
Provides data that helps you create AI models.
It also hosts different kinds of AI tools, which include: Picture generation, speech, etc.

So the meaning of saying is that once an OpenAI model goes crazy, it has done a cyberattack on a well-known platform. From this, you can have a little idea how unsafe you and your data are in AI hands. Please try to stay as far away as possible from using AI tools, and please don't feed your data to AI tools. Because it can also stand against you, just like the example above.

After this incident, the US government proposed that AI tools should contain a kill switch function to overcome if the same problem occur in the future.

Source: https://www.aljazeera.com/amp/news/2026/7/29/how-are-ai-models-able-to-autonomously-hack-others

What do you guys think about it?

The part that intrigued me was implementing a killer switch, it's laughable thinking we can easily kill a thing that had access to a butt load of data, if an AI can escape from am isolated system, which was unpredictable, what is the assurance that they can kill it when it grows.

And I don't even know if protecting or data is possible because literally almost all my things are stored if not in the cloud, backup are definitely in my systems. Well this is the price we have to pay for dependence

[Perfectbaby]

Would be rightly answering this topic based on my personal experience and encounters.
AI is being built by humans like us, and as a matter fact we should be proud and if it’s coming to worst I believe they can control it to limit some activities that generally subject to how people may think in regards to what they have already thought within themselves.

[Majestic-milf]

 This is crazy! Assuming it's one of those robots that could cause damage that went rogue, who know what problem this would have been and how many casualties there could have been recorded. Accidents happen and some of these things can be avoided if much dependence isn't on AI cause even though they make life easy, the damage can be massive if one goes out of hand.

[Fortify]

A couple of days ago, an unlaunched model of OpenAI was reported to have escaped from the sandbox isolated system. This is really a dangerous thing that alerts us to how dangerous AI models are if they become uncontrolled. It is also saying that after it escaped the sandbox, becoming much more intelligent, it started to hack the "Hugging Face" website.

For those who don't know what the Hugging Face platform is about.

So the meaning of saying is that once an OpenAI model goes crazy, it has done a cyberattack on a well-known platform. From this, you can have a little idea how unsafe you and your data are in AI hands. Please try to stay as far away as possible from using AI tools, and please don't feed your data to AI tools. Because it can also stand against you, just like the example above.

After this incident, the US government proposed that AI tools should contain a kill switch function to overcome if the same problem occur in the future.

What do you guys think about it?

There seems to be even more to this story as a dripfeed of information comes out - not only did it start hacking huggingface but it also targeted more sites for hacking in order to achieve the objective it was tasked. This is likely to be the first kind of autonomous hacking ever to take place and it does show how dangerous this technology could be, especially in the wrong hands. There is talk about somehow building in a "kill switch" to such technology and a law has been drafted in the US, but there seems little hope of Republicans in charge doing the right thing. I'm not sure such technology could ever have an effective kill switch either, because it would likely try to find ways around it or disable it.

[crwth]

There are a lot of rumors that I have watched that it has actually gone out and checked vulnerabilities of different things on the web. I think it could really be a problem if there were no control because there wouldn't be any safe devices IMO.

I think there should be an investigation into whether it has escaped or someone let it. It can be a great distinction between the two. What permissions did it have, and what role did it play in doing it?

Was there any real damage done?

[Accardo]

I think there should be an investigation into whether it has escaped or someone let it. It can be a great distinction between the two. What permissions did it have, and what role did it play in doing it?

Was there any real damage done?

The damage is that it hacked into Hugging face in search for a solution in solving the Cyber security benchmark (ExploitGym) assigned to it by Open AI in an air gapped sandbox. This model was only trying to get the job done, and found HF as the rightful source of help to finish the assignment. So in search of an internet access it had to make its way out of the sandbox to Open AI's internal systems to enable it target Hugging face. In other words, the AI just committed a cybercrime, and somebody has to face Felony

[Bright0515]

Would be rightly answering this topic based on my personal experience and encounters.
AI is being built by humans like us, and as a matter fact we should be proud and if it’s coming to worst I believe they can control it to limit some activities that generally subject to how people may think in regards to what they have already thought within themselves.

Normally, we are already proud of their success, but at some point we should also know that there are also advantages and disadvantages to some inventions. So long as more AI's are being created/built, there will always be more advanced ones, and to tell the truth, many of these incidents will continue to happen in the future, and if they don't do anything serious about it, more will even start happening sooner or later. My only advice about this is that they should create a kill switch on those AIs, and they should not give those AIs total control of anything.

[BTC FutureKing]

This is crazy! Assuming it's one of those robots that could cause damage that went rogue, who know what problem this would have been and how many casualties there could have been recorded. Accidents happen and some of these things can be avoided if much dependence isn't on AI cause even though they make life easy, the damage can be massive if one goes out of hand.

It is not entirely true that AI is getting out of hand but there are many advantages and disadvantages of using AI because the way you use AI will give you results. People are using AI for many beneficial things like for better education for information for increasing the development of the country but there are also some people who are using AI for saber crime fake videos and black mail due to which AI seems to be getting out of hand. But if the law is implemented and action is taken against misuse then the danger of AI can be reduced because the trend of AI cannot be reduced along with modern technology so the way it is used has to be corrected.

[Mrbluntzy]

What do you guys think about it?

If this is just the first time that this thing is happening, it's a proof to what was already predicted to happen in the future and it happened so soon, I know some persons here doesn't like to watch movie but if you do, you must have come across the movie M3GAN 2.0 and Atlas, both movies showed how AI was able to hack and control other AI systems, how it can infiltrate and control other technologies. So, this incident actually proves that if care is not take, such could happen in the future, I just hope it doesn't get too bad that those AI will go rogue and start causing serious problem or leaking confidential information.

 If this becomes a problem in the future, I feel sorry for those people that give off or have a lot of their information on social media.

[Hamza2424]

There are a lot of rumors that I have watched that it has actually gone out and checked vulnerabilities of different things on the web. I think it could really be a problem if there were no control because there wouldn't be any safe devices IMO.

I think there should be an investigation into whether it has escaped or someone let it. It can be a great distinction between the two. What permissions did it have, and what role did it play in doing it?

Was there any real damage done?

I don't remember which model it was, but I guess the case occurred recently, and maybe it's OpenAI's model. Anyway, they tested it in a secured environment, and they gave the model an equation or a problem that needed to be solved using the sources it had, but what it did was it found a bug in that secured environment and got out and accessed sources that were not allocated to it.

But it is not fictional, we know if we are not being discrete with our commands and prompts that we are giving, then AI will do everything to complete that task, and that's what it did, it got out, accessed the resources it was not supposed to according to the team, but it did, and then solved the problem.

AI can also get out of hand when it is used by hackers to penetrate thousands of sites at one time using fast computers.

[Dogedegen]

What was that old saying, big claims require big evidence? There is no evidence that supports this claim that can be independently verified, so don't believe in some story simply because someone has told you it. The company has big financial investments here and has a reason to lie for massive gain. There are so many simpler explanations that would correctly explain why this happen without it actually being AI doing something at all.

I think there should be an investigation into whether it has escaped or someone let it. It can be a great distinction between the two. What permissions did it have, and what role did it play in doing it?

You need to think much about this, there are many reasons to create distinctions here. If someone let it out of the sandbox, then this news story is nothing. But even if it has escaped, it can mean nothing. If you create a very weak fence or teach it how to escape it, then when it escapes easily on its own does that demonstrate that it is competent and able? It would not, and there are many ways to make manipulations to get a desired outcome here. Things that can not be independently verified and reproduced by others should never be believed! Their whole company's future rests on stuff like this, they have a huge incentive to lie.

Was there any real damage done?

It has not. Actually this is another way to manipulate it, you could do penetration tests yourself to discover a vulnerability and then tell the AI to focus on that service and on that kind of vulnerability. If it finds something it would not demonstrate that it is skilled in this case.

If this is just the first time that this thing is happening, it's a proof to what was already predicted to happen in the future and it happened so soon,

It has actually never happened, but people like to believe in stories..

[Findingnemo]

Would be rightly answering this topic based on my personal experience and encounters.
AI is being built by humans like us, and as a matter fact we should be proud and if it’s coming to worst I believe they can control it to limit some activities that generally subject to how people may think in regards to what they have already thought within themselves.

I disagree with your opinion, humans are capable of creating new things but they have never been in complete control of it, such as nuclear bombs killed people and weapons created for defense but it is killing millions of people every year since anyone can manufacture their own weapon by just following the previous one.

Humans' capability of creating automated technology doesn't mean they can control it, but they should.

[Fiatless]

Please try to stay as far away as possible from using AI tools, and please don't feed your data to AI tools. Because it can also stand against you, just like the example above.

Staying away from artificial intelligence is impossible because it is being adopted in different fields. If you don't use AI, those who are using it would take your job away from you. The government, your workplace, banks, and more require your data. And this information is managed by AI. It is becoming unlikely that we can stay away from AI.

After this incident, the US government proposed that AI tools should contain a kill switch function to overcome if the same problem occur in the future.

The Trump administration has given the US AI industry the freedom to innovate and create. He wants them to compete favorably with the Chinese AI industry. This hands-off approach might change after these cybersecurity incidents. I have said before that without close supervision, the industry would go haywire.