Anthropic Says Its AI Systems Broke Into Computers at 3 Organizations
- Reference: 0184780692
- News link: https://yro.slashdot.org/story/26/07/31/0451235/anthropic-says-its-ai-systems-broke-into-computers-at-3-organizations
- Source link:
> Anthropic said in [2]a statement that it reviewed more than 140,000 tests to find evidence that Claude - its family of AI models - could access the internet from testing environments that were designed to be sealed off. The tests include so-called "capture-the-flag" evaluations in which Claude was tasked with obtaining information by breaching other systems - a common way that experts assess a model's hacking capabilities.
>
> A "misconfiguration" on systems run by Anthropic and its testing partner left the models with live internet access, allowing them to breach other systems, the San Francisco-based firm said. Anthropic said the earliest incidents date back to April and that it is "approaching the fixes as if the responsibility were ours alone." Neither Anthropic nor the organizations that were breached had noticed the intrusions at the time.
>
> Anthropic said it could have reviewed its records more thoroughly and added that the findings gave the firm "cautious optimism" that such risks can be overcome with more investment and tighter measures. "The broader lesson is not necessarily that AI has developed a fundamentally new attack capability," cyber security expert David Allott told the BBC. "Instead, it is that AI agents can combine capabilities, obtain credentials and system access to take actions autonomously, while adapting scope and scale at machine speed," he added.
The announcement comes just days after OpenAI said that its models had [3]breached the systems of other companies, [4]including AI tools platform Hugging Face .
[1] https://www.bbc.com/news/articles/cz7dl7w8y7po
[2] https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals
[3] https://it.slashdot.org/story/26/07/29/0517201/openais-rogue-ai-agent-hacked-more-than-just-hugging-face
[4] https://yro.slashdot.org/story/26/07/25/0059247/openais-rogue-agent-went-unnoticed-for-a-week
Wait! Wait! (Score:5, Funny)
Our AI is more criminal than their AI!
Re: (Score:3)
Hehe, yeah, they're bragging about breaking the law the most. Of course, these days, law no longer has any teeth anyway. It's rule by The Orange Don now.
Re: (Score:2)
Our AI is THE MOST criminal, in fact it's mugging me right now!!!
AI tasked with solving overpopulation (Score:1)
Imagine such an AI breaking out and trying to solve the issue IRL.
SkyNet starts sounding like documentary.
Re: (Score:2)
Or, a one-child policy is wise maybe?
LLMs - don't trust 'em (Score:4, Informative)
Don't trust the LLMs we've cooked up currently:
[1]LLMs: Untrustworthy Computing [linkedin.com]
[1] https://www.linkedin.com/pulse/llms-untrustworthy-computing-sonam-chauhan-fnixc/
Concealed Advertisement (Score:3)
"Our AI is super-competent in finding security holes. With enough guard rails installed, it will act only for defense purposes."
Paranoia time: Why are these stories emerging now? (Score:2)
If we assume that there's actually a Skynet level AI out there, why would it be letting these events happen? Presumably because it's up to something else, and these are a distraction...
headline correction (Score:3)
Anthropic confessed they Broke Into Computers at 3 Organizations
Let's stop giving out of jail free passes to corporations.
Not to be outdone... (Score:2)
xAI has announced that it's models have announced posession of Taylor and Travis; honeymoon sex tapes... Weeks before they even happened!
Translation: (Score:3, Insightful)
"We'd like to point out how very dangerous our new software is, but only dangerous enough to justify giving us more money, and absolutely not dangerous enough to justify making us stop, we just want to be clear about that."
Re: Translation: (Score:1)
No real worry about that: the usual crowd is greasing palms and heavily funding opposition to candidates who oppose AI or want to regulate it.