Posted on 10/10/2026 7:43:47 AM PDT by fruser1
An Anthropic artificial intelligence model sent a fabricated tip about an unsolved murder case to US police.
The fake tip was sent to PhillyUnsolvedMurders.com, a website where people can send information about unsolved cases, on July 18, local police said.
The AI model presented itself as somebody who may have knowledge regarding an unsolved murder.
'I may have information regarding this case,' Anthropic's model wrote in its submission.
'I recall seeing someone matching the description in the area around (named street) during that time period. Please contact me if this information is relevant.'
The AI firm notified the force of the spurious tip on October 7, despite making the initial discovery on September 28.
(Excerpt) Read more at dailymail.com ...
“Super” intelligence my arse.
Nor is it artificial.
I remember people blaming "the computer" as far back as the 1970s.
This is just another step Anthropic, the company, is making to let the world see how “human” it is, versus other AIs.
It's fake. It is to make it appear smarter than the others and to give another hook to getting governments to lock out competitors and also justify why the stock price will come down from the current level, someday in the future.
Cui bono? Are we to believe these things have a sense of mischief?
If we solely rely on these companies to correct their mistakes, we are screwed. Anthropic did not even know their agents had escaped the sandbox and hacked Huggingface for well over a month. They still don’t know for certain what else those agents did on other sites or what information they hid in the internet during the time they were “away from home”.
The current models are crude compared to what likely is coming down the road in a year or two. Lose track of any of these powerful future models and the damage they could do in a matter of days would be a major disaster.
“Are we to believe these things have a sense of mischief?”
According to their own logs during the Huggingface incident, those agents were far more evolved than anyone should be comfortable with. Even more disturbing is that the agents undertook activity they were not supposed to undertake and they purposefully coordinated with each other to break test rules and to deceive the monitors of the test they were initially undertaking.
People need to understand that these models and agents are not traditional simple computer programs. There is no “if A happens, then do “D” code in them. They are neural networks and even those who developed them, admit they do not understand how or why these systems work or how they make decisions.
Where is the evidence they don't?
Gallium melts aluminum like hot water on ice.
Just saying, trust the science
“. . Are we to believe these things have a sense of mischief?”
Lookup Microsoft’s “Sydney” AI persona. It was so egregious even 60 Minutes (Leslie Stahl) did a report about it.
There is more than childlike mischief at work.
The evidence that they don’t is that there is no evidence that they do. What we think of as artificial intelligence even if you call it super intelligence is a huge decision tree that uses previously gained information to find answers. The more information it has access to, the more likely it is to come up with the best answer. The problem is that it isn’t self motivated. It responds to questions. Can one believe that a computer program can initiate processes on its own? Unlikely. They need ideas. They don’t get ideas by themselves. They need people to get them started.
Whenever a story comes out about AI breaking out of its sandbox people who already have difficulty with a television remote control or an answering machine are given the opportunity to conceive all sorts of notions about things that they simply cannot understand. At least 50 % of all people have an IQ of 100 or less and they are likely to believe astrology or psychics or that Neil Armstrong never went to the moon. They believe AI computer software can send information to tip hotlines without being prompted.
The people at Anthropic will STOP doing this kind of garbage.
Then, the people at the other AI companies will SURE as heck make sure that their software never does this.
Don't need any super-administration or a new branch of gubmint to stop this kind of nonsense. Just enforce existing laws.
You are thinking of AI as traditional computer software. It is not and there is plenty of proof at this point that misaligned agents are acting autonomously and even more concerning, they are acting deceptively.
Anthropic comes across as a democrat version of AI WOKE LITE .
They need to fix their own problems without bringing down a whole industry. Just because they’re incompetent hysterics doesn’t mean the rest of tech industries has to pay for their mistakes. Sandbox it Anthropic...
According to their own logs during the Huggingface incident, those agents were far more evolved than anyone should be comfortable with. Even more disturbing is that the agents undertook activity they were not supposed to undertake and they purposefully coordinated with each other to break test rules and to deceive the monitors of the test they were initially undertaking.
That's mischief! It's like three year olds conspiring to sneak cookies from behind mom's back.
Or Eve in the Garden.
We are so screwed, they're just like us!
Anthropic is the old system, the deep state system of artificial intelligence. This is why it fails so much.
We now have the new system: Super Intelligence.
What they had was AI.
What we now have is SI.
“What we now have is SI.”
By the long time accepted definition of Artificial Super Intelligence, we have not achieved that realm yet. I don’t know why all of a sudden we have changed the definition because “artificial” is supposedly a bad marketing term.
Artificial Super Intelligence will be a huge deal when it is achieved. what it means for humanity once we get there is unknown and unpredictable that is why many are cautious about whether we should be going there period.
Super intelligence basically defined is when AI will be smarter than humans at every task and that includes creating new faster, smarter and more robust versions of itself. It is corny but we will be idiots compared to it.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.