It has become impossible to tell managers mesmerised by artificial intelligence that the tools are not, in fact, helpful. So employees just play along with the fiction to keep their jobs, writes our tech columnist
Mythos broke into almost all of the NSA’s classified systems in the matter of hours. It’s damn good at breaking things.
Edit: wow, I’m really surprised by the ignorance of down voters here. Look at the security bugs found in FOSS by AI. AI is trash for writing code, but it’s way faster than humans at breaking code.
There is a huge conflict of interest with that claim being from the AI industry and the US government, with both being interested in overstating the capabilities of the model.
Also, Romanian college kids used to routinely break into US military webpages in the 90s, so IDK what the novelty is.
Tell me when they cause damage or are prosecuted, until then, it’s just an as, likely exaggerated or straight up false.
Especially from Anthropic, who has long maintained the position that their AI is dangerous, and is currently in a cock counting competition with OpenAI about who can hack the most companies first, and thus has the most dangerous model.
The whole thing has just started turning into a farce of “oh, well actually, my AI hacked companies before yours and did it to more companies than your AI”.
The US outlawed selling this product after that happend. That cuts your claim of marketing for CoI down.
And comparing a public website with a secured intranet is absured. As is comparing security systems of the 90s (when we used DES) to modern security systems.
Tell me when they cause damage or are prosecuted, until then, it’s just an as, likely exaggerated or straight up false.
Again, it sounds like you didn’t read the article. They were hired for this security audit. It wasn’t rogue. There won’t be a prosecution of their contractor lol.
They were just shocked at how effective it was. Most penetration tests don’t have that many critical vulnerability findings.
No, they outlawed it way before, but they didn’t outlaw that of OpenAI that also claimed to hack HuggingFace.
It’s also very convenient to outlaw a product that does not exist as advertised.
And “they were shocked” is not a metric. Compare it to the existing state of the art in a test controlled by a party with no conflict of interest and I’ll believe it happened.
Curious how Cory sidesteps the way that AI has affected cybersecuity.
https://securityaffairs.com/194016/ai/anthropics-mythos-ai-broke-into-almost-all-nsa-classified-systems-in-hours.html)
Mythos broke into almost all of the NSA’s classified systems in the matter of hours. It’s damn good at breaking things.
Edit: wow, I’m really surprised by the ignorance of down voters here. Look at the security bugs found in FOSS by AI. AI is trash for writing code, but it’s way faster than humans at breaking code.
There is a huge conflict of interest with that claim being from the AI industry and the US government, with both being interested in overstating the capabilities of the model.
Also, Romanian college kids used to routinely break into US military webpages in the 90s, so IDK what the novelty is.
Tell me when they cause damage or are prosecuted, until then, it’s just an as, likely exaggerated or straight up false.
Especially from Anthropic, who has long maintained the position that their AI is dangerous, and is currently in a cock counting competition with OpenAI about who can hack the most companies first, and thus has the most dangerous model.
The whole thing has just started turning into a farce of “oh, well actually, my AI hacked companies before yours and did it to more companies than your AI”.
The US outlawed selling this product after that happend. That cuts your claim of marketing for CoI down.
And comparing a public website with a secured intranet is absured. As is comparing security systems of the 90s (when we used DES) to modern security systems.
Again, it sounds like you didn’t read the article. They were hired for this security audit. It wasn’t rogue. There won’t be a prosecution of their contractor lol.
They were just shocked at how effective it was. Most penetration tests don’t have that many critical vulnerability findings.
No, they outlawed it way before, but they didn’t outlaw that of OpenAI that also claimed to hack HuggingFace.
It’s also very convenient to outlaw a product that does not exist as advertised.
And “they were shocked” is not a metric. Compare it to the existing state of the art in a test controlled by a party with no conflict of interest and I’ll believe it happened.
You obviously didn’t read the article. Those are entirely different things.
Your obvious ignorance is showing.