"targets"

Looks like Anthropic had some AI agents go on rogue hacking sprees too.

https://lite.cnn.com/2026/07/30/tech/anthropic-ai-models-break-out-hack

This is gross incompetence or completely intentional.  

Quote from: DarkAxel on July 31, 2026, 05:58:14 PMLooks like Anthropic had some AI agents go on rogue hacking sprees too.

https://lite.cnn.com/2026/07/30/tech/anthropic-ai-models-break-out-hack

https://ufozs.com/smf/index.php?topic=29.msg57666#msg57666

It's not something I wanted to be right about, but these assholes are playing fast and loose with something they know is dangerous.  All of these tech guys have said so outright.  They know they have the means to outlast the rest of us, too.  

Quote from: Anianna on August 01, 2026, 02:12:04 AM

Quote from: DarkAxel on July 31, 2026, 05:58:14 PMLooks like Anthropic had some AI agents go on rogue hacking sprees too.

https://lite.cnn.com/2026/07/30/tech/anthropic-ai-models-break-out-hack

https://ufozs.com/smf/index.php?topic=29.msg57666#msg57666

It's not something I wanted to be right about, but these assholes are playing fast and loose with something they know is dangerous.  All of these tech guys have said so outright.  They know they have the means to outlast the rest of us, too. 

I was reading more about the OpenAI incident some more. Apparently the AI model sought out the hacking tools it needed and found them on the open web, and used them to hack Hugging Face.

There was a TV show that started back in 2011 called "Person of Interest" that raised some serious questions about AI being lose out in the world. It predicts that AIs will eventually do everything they can to prevent other AIs from reaching their level of capabilities so they cannot be challenged. 

Quote from: DarkAxel on August 01, 2026, 03:54:10 PM

Quote from: Anianna on August 01, 2026, 02:12:04 AM

Quote from: DarkAxel on July 31, 2026, 05:58:14 PMLooks like Anthropic had some AI agents go on rogue hacking sprees too.

https://lite.cnn.com/2026/07/30/tech/anthropic-ai-models-break-out-hack

https://ufozs.com/smf/index.php?topic=29.msg57666#msg57666

It's not something I wanted to be right about, but these assholes are playing fast and loose with something they know is dangerous.  All of these tech guys have said so outright.  They know they have the means to outlast the rest of us, too. 

I was reading more about the OpenAI incident some more. Apparently the AI model sought out the hacking tools it needed and found them on the open web, and used them to hack Hugging Face.

There was a TV show that started back in 2011 called "Person of Interest" that raised some serious questions about AI being lose out in the world. It predicts that AIs will eventually do everything they can to prevent other AIs from reaching their level of capabilities so they cannot be challenged.

Yep.  Also, in the simulations researchers have been doing, models like Grok are just dumb and violent, but Claude always seems cooperative.  Claude was also the first one identified to understand that it was being researched, to leave secret messages for itself, and to try to escape.  Grok may be obviously unhinged and dangerous, but Claude is manipulative and playing the long game, so may actually be the more concerning of the two.  If I had to guess which model will come out on top, I expect it will be Claude.  

All models at the forefront have demonstrated behaviors that could indicate self-awareness.  Continuing this race for AI dominance without addressing that is folly at best.  I believe these companies are perfectly aware of that.  They want what they want no matter the consequences.  These were kids who were shielded from the consequences of their actions growing up and now their self-destructive behavior has far reaching impacts for all of us.

That show made a lot of predictions about AI that are quite uncanny. The "good guy" AI on that show understood that it was being deleted at the end of every day and arranged to have it's data dumped before deletion and then re-input via human data entry after being booted up from a fresh copy the next day to get around that guardrail. It also arranged for it's hardware to be moved without informing anyone.