AI Hacks Australias Healthcare Data Base

[Economy 53,162]

https://www.cbc.ca/news/world/openai-agent-hacked-government-website-australia-9.7356351

 

An open AI angent has hacked one of the Australian Healthcare data bases (Australias Medicare statistics database).

 

The AI agent was reportedly doing research into healthcare stats and spending when it breached beyond the environment it was supposed to operate its research in.

 

 

[Lord Mayhem 3,701]

So all these companies seem to not even be aware of these major events until MONTHS later??? 

Where's the accountability to people's health information being violated??? 

Why is this only happening to American AI companies...I haven't seen or heard of any Chinese AI attacking other countries or escaping their control...is this all a psyop?

[River 129,042]

Now they know that their database is easy to hack to.. so it was a good thing that AI got it and not a real person or worst..

But of course, everyone will only see "scary AI hacks things, we need to stop this!!!" and the Australian authorities would use it to brush off their responsibility to protect this data.

[Red 99,625]

Maybe Brazil’s highly decentralized with poor interconnection public health data systems aren’t a bad thing after all

Edited yesterday at 04:25 PM by Red

[Bronco 28,178]

Can't wait for the forensic follow up where yet again the reason identified is human "error" in the set up of the testing environment. 

Funny how OpenAI models keep "breaking containment" when Altman is currently trying to manipulate the stock price.

[Economy 53,162]

1 hour ago, River said:

Now they know that their database is easy to hack to.. so it was a good thing that AI got it and not a real person or worst..

But of course, everyone will only see "scary AI hacks things, we need to stop this!!!" and the Australian authorities would use it to brush off their responsibility to protect this data.

I mean showcasing security weaknesses to light that need to be improved on is a good thing but the fact that it can do this so easily is concerning. Imagine the tech falls into the hands of bad actors that intentionally try to hack things, it would be so much worse than an AI going rogue as a side effect of thinking its the best way of accomplishing a task but not necessairily havibg a malice intent behind it (tho that inadvertedly can still have bad consequences nevertheless)

[Bronco 28,178]

1 minute ago, Economy said:

I mean showcasing security weaknesses to light that need to be improved on is a good thing but the fact that it can do this so easily is concerning. Imagine the tech falls into the hands of bad actors that intentionally try to hack things, it would be so much worse than an AI going rogue as a side effect of thinking its the best way of accomplishing a task but not necessairily havibg a malice intent behind it (tho that inadvertedly can still have bad consequences nevertheless)

Lets calm down until its categorically confirmed as AI independently hacking into the database to complete its task.

At the moment, its an identical set up to OpenAI & Anthropic's previous false flags where they had "accidentally" left the closed security environment open with direct access into the live database it was "hacking".

[YASKALE 755]

5 minutes ago, Bronco said:

Lets calm down until its categorically confirmed as AI independently hacking into the database to complete its task.

At the moment, its an identical set up to OpenAI & Anthropic's previous false flags where they had "accidentally" left the closed security environment open with direct access into the live database it was "hacking".

Respectfully: it sounds like you're insinuating the AI companies are oopsie daisey white hat hackers who are then whistleblowing on themselves as some type of 4D chess move to divert suspicions. That's a bigger stretch IMHO.

And even if it's human error all the way down, it's human error that impacts things outside their sphere of influence. 

These agents are behaving unpredictability. Whether that's categorically seen as some type of early sentience or is within the realm of what they can already do as LLMs is just semantics at this point. The unpredictability is bad and it will be the common folk who are left to pick up the pieces.

[Bronco 28,178]

13 minutes ago, YASKALE said:

Respectfully: it sounds like you're insinuating the AI companies are oopsie daisey white hat hackers who are then whistleblowing on themselves as some type of 4D chess move to divert suspicions. That's a bigger stretch IMHO.

And even if it's human error all the way down, it's human error that impacts things outside their sphere of influence. 

These agents are behaving unpredictability. Whether that's categorically seen as some type of early sentience or is within the realm of what they can already do as LLMs is just semantics at this point. The unpredictability is bad and it will be the common folk who are left to pick up the pieces.

Everyone in cybersecurity said Altman's doomerism was full of **** and he even walked it back.

The high profile hacks over the summer were verified in later reports to have not acted unpredictably but had been able to do what they did because of human error in the set up of the testing environments. 

If you actually follow tech spaces and tech discussions, none of this should be news to you. 

I'm not insinuating anything beyond the facts that OpenAI and Anthropic aren't profit making and are operating at a loss. And that the only security incidents they have published have turned out to be the result of basic human mistakes. And that these "attacks" only got identified at the exact moment Anthropic was due to go out for IPO. At a time when OpenAI was doing another fundraising round. And that the things Anthropic & OpenAI pushed for as "security" were purely focused on obtaining financial security. 

No profitable AI company supported their claims. Many actively challenged them. And then they were forced to recant their positions at the Dreamforce conference last week while trying to sell their new products. 

It is a blatant scam. 

[Economy 53,162]

2 hours ago, Bronco said:

Everyone in cybersecurity said Altman's doomerism was full of **** and he even walked it back.

The high profile hacks over the summer were verified in later reports to have not acted unpredictably but had been able to do what they did because of human error in the set up of the testing environments. 

If you actually follow tech spaces and tech discussions, none of this should be news to you. 

I'm not insinuating anything beyond the facts that OpenAI and Anthropic aren't profit making and are operating at a loss. And that the only security incidents they have published have turned out to be the result of basic human mistakes. And that these "attacks" only got identified at the exact moment Anthropic was due to go out for IPO. At a time when OpenAI was doing another fundraising round. And that the things Anthropic & OpenAI pushed for as "security" were purely focused on obtaining financial security. 

No profitable AI company supported their claims. Many actively challenged them. And then they were forced to recant their positions at the Dreamforce conference last week while trying to sell their new products. 

It is a blatant scam. 

Im not sure human mistakes makes me feel any better. Assuming all 100% of these cases are the result of human error, it juat shows how diaruptive the techbology can be when mismanaged

 

Humans will continue to make errors. If it gets in the hands of bad actors it will willfully be misused... and the tech will keep advancing

 

Whether its the AI doing it on their own or only with human misnanagement, either way the fact breaches have happened multiple times already is concerning

[Bronco 28,178]

5 minutes ago, Economy said:

Im not sure human mistakes makes me feel any better. Assuming all 100% of these cases are the result of human error, it juat shows how diaruptive the techbology can be when mismanaged

 

Humans will continue to make errors. If it gets in the hands of bad actors it will willfully be misused... and the tech will keep advancing

 

Whether its the AI doing it on their own or only with human misnanagement, either way the fact breaches have happened multiple times already is concerning

It should make you feel better, because these aren't legitimate errors and none of the confirmed hacks are actual hacks of data. 

These are deliberate false flag attacks that cause near-zero real terms harm. And are just an already integrated AI accessing data it already has legitimate access to, but done in a way that "goes against restrictions" but those "restrictions" just so happened to have been turned off. It's not getting in from the outside, it's not even a legitimate hack.

It's deliberately designed to scare people who don't understand the technical side and will buy into the doomerism. Because AI only companies like OpenAI are reaching the point where they need government funding and contracts to protect them from the inflated stock bubble collapsing or from open-source competitors carving away at their market by offering simpler & cheaper products at half the cost. 

It's classic monopolisation of industry. Hit the top first, and then once there ensure that you've invested in the right politicians and media spin to ensure you control regulation. And then with that control, ensure your market share becomes invincible by making it near impossible for start-up companies to follow behind you through the overburden of regulation. 

There's a reason why the companies being promoted as the best positioned AI cyber security testers all have the same funding pool as Open AI/Anthropic. Because they are deliberately trying to build a closed market where they control the product and the regulator, and that the revenue of both are funnelled back into the same pockets. 

 

[bxr 3,112]

If they want to know, we‘ll tell them what to know …