BS Summary: This video contains 26 faulty reasoning types, including Appeal to Authority, Post Hoc (False Cause), and Availability Heuristic, with Negativity Bias as the most egregious example at 29.6% saturation with 120 hits. Analysis detected 1,106 faulty-reasoning hits from 405 analyzed words, generating a BS Score of 80.6% and a BS Rank of 91% (2,467 of 26,005 videos). This video is worse (more manipulative) than 90.50% of the video peer group.
Now to a growing concern about the evolution of artificial intelligence.
AI giant Anthropic says one of its models hacked into three other organizations during testing.
Now that word coming just days after OpenAI said one of its own models went rogue.
Technology reporter Mike D'Abrosca has the details.
>> So Anthropic says it was running some tests on its AI agents in recent months and in some of those tests the agents broke free of their testing environments and hacked into three separate organizations.
We know the names of these models Opus 4.7 and Mythos 5.
There was also a third model that was implicated here but Anthropic isn't naming it just saying it's one of their upcoming models.
All this stuff was being run without the normal guardrails in place.
This is important.
These are not models that you or I could really use.
What we don't know is who was hacked.
Anthropic isn't naming those organizations.
Now if this sounds familiar it's because something similar happened with OpenAI just a few weeks ago.
Anthropic is trying to distance itself from OpenAI but that's no surprise since Anthropic often casts itself as the more responsible alternative to the much larger OpenAI.
They basically say this was a misconfiguration problem which allowed these models to access the internet when they shouldn't have had that access.
Someone basically failed to flip a switch.
Essentially they also reacted better than their competitor according to Anthropic saying they launched this preemptive internal review after the OpenAI incident from a few weeks ago which led to them discovering these breaches.
So the big question, should people be concerned?
Well, to be clear this is not something like the Terminator or Skynet coming online.
These are not models deciding to hack into things on their own.
Remember, they're taking a test.
They're being told to find a solution to a problem.
The internet has the answer to a lot of problems and these models are pretty powerful.
Powerful.
So while it's scary to hear phrases like goes rogue, gains access to the internet, let's remember that these are machines that had been given a problem to solve and ultimately they did.
However, they did that in a way that highlighted how lax some of the security policies at both of these companies may be.
Back to you.
>> All right, Mike Dobuski.
Thank you.
Analysis
Hover over highlighted words in the article to view the associated bias or fallacy analysis.