Paul Stenhouse: Anthropic's new model could be a major cybersecurity threat and a survey reveals parents' concerns about their children's online safety

The next level of AI is about to be released, and there are major concerns  

The models just keep getting better and better and now Anthropic, the maker of Claude, says its software engineering capability is now so advanced that it's finding thousands of vulnerabilities in software that no one knew existed.  

Anthropic says "the fallout for economies, public safety, and national security could be severe". Anthropic has shared its new model Mythos with an alliance of cybersecurity companies as well as another 40 organizations. The CEO of Cisco, the maker of networking equipment, says AI has reached a level where critical infrastructure needs to be protected. US Treasury Secretary Scott Bessent met with critical US banks in the wake of the model's preview release.  

 

Three quarters of parents fear their child could not make safe online privacy choices  

The survey by the UK's data watchdog shows we need to be teaching kids about online safety just like we would road safety, because it's an essential life skill.  

• 35% of parents think their child would share personal information in exchange for game tokens. 
• 22% of children had shared health details with AI tools. 
• 24% of children had shared their real name or address online. 
• and 21% of parents had never discussed online privacy with their children. 

 

LISTEN ABOVE 

See omnystudio.com/listener for privacy information.

Episode Transcript

Speaker 1 (00:07):
You're listening to the Saturday Morning with Jack team podcast
from News Talks at be.

Speaker 2 (00:12):
The next level of AI is about to be released
and there are big, big concerns. Our textbit poor Stenhouse
is here with the details. Good a Paul, Yeah, Good
morning Jack.

Speaker 3 (00:21):
It's hard do sometimes know with this AI stuff when
we hear these developments, how much of it is sort
of the marketing height or the or you know, trying
to hype the investors. Do you give us some more
money and we need more money for infrastructure? And how
much of it is actually real? But I will say
that this model is coming from anthropic the ore, the
makers of Clawed and there I think it was their

(00:43):
CEO said that the fallout for economies, public safety, and
national security could be severe. So that kind of goes
beyond marketing hype, I think at that point. So this
is probably one we need to pay attention to.

Speaker 2 (00:59):
I think that the way they've been discussing it, and
I love the stuff as you do, but I think
the way they've been discussing it suggests that they're seriously
concerned and about liability, Like they aren't talking it up
in a big way because they want to make money
out of it necessarily, though of course they do. It's
more that they're concerned this is so powerful that if
they release it for everyone overnight, the things that people

(01:19):
could potentially do with it would leave them liable for
lawsuits that would bankrupt the company, you know, I mean yeah, I.

Speaker 3 (01:26):
Mean when you start talking to some of the things
here that it could be finding. Right, So this latest
model mythists is able to find. They're basically saying that
it is probably in the top one percent of software
engineers around. They're saying that it is in the top
one percent of cybersecurity professionals around. And so it is
able to find. It has found thousands, thousands of vulnerabilities

(01:48):
and software that we assumingly use day in and day out.
Some of the vulnerabilities that found have been there for
twenty seven years. No one knew they were there, even
the companies that have been testing it and know all
about the software. So this is like very powerful stuff.
And I think the thing that is probably most alarming
is it's not like it could just you know, create

(02:09):
a really cool app, you know that's for good. They're
really concerned about the bad And the thing that kind
of got me you know, set my alarm bells off.
Was the US Treasury Secretary. He actually in the wake
of this, a bunch of the US banks happened to
be in Washington for some lobbying type of things. But
he met with all of these critical banks. And you

(02:31):
can imagine, or I guess you don't want to imagine,
but you could be a time where imagine if they
could get into the kind of the critical big US
banks and they could start messing with transactions or messing with,
you know, the flow of money like that would just
upend them. So that would it would be and so
that wouldn't just be a lawsuit and some liability, that

(02:54):
would be something quite different.

Speaker 2 (02:56):
And this is the concern, right, So they've basically tested
this model. They've asked this model to find cybersecurity vulnerabilities
and some of the most important software in the world.
And so this is stuff like internet browsers. It's the
little bits of software that none of us even know
is being used. So the open source stuff that gets

(03:17):
used in all manner of software that you use every
day without even kind of being aware that you're using
software at a time. And you know, there are examples
that I've read about where, for example, a piece of
software has been checked five million times by human cybersecurity experts,
coding experts, and artificial intelligence models and has never been

(03:40):
found to have a problem. And now they've put this
mythos model on it and it's immediately found massive vulnerabilities
that could be catastrophic. And it's this kind of example replicated, however,
many times over with However, many companies, however, many forms
of software that is really worrying.

Speaker 3 (03:56):
Yes, And one of the ones that also got my
kind of hairs on the back of my neck up
was Cisco. So the CEO of Cisco, and Cisco makes
networking kind of hardware, right, but obviously behind the hardware
is also some software. Now, when you've got the CEO
of Cisco coming and saying that AI has reached a
level where critical infrastructure needs to be protected, as Cisco,

(04:17):
who provides the critical infrastructure for many digital businesses and
huge networks, you have to think we might be on that,
we might be at a time where we might need
to pump the brakes.

Speaker 2 (04:30):
Yeah yeah, so interesting A. So one of the steps
this company has taken is they've got an alliance of
cybersecurity companies and like another three or four dozen organizations,
and they've given them like one hundred million dollars worth
of tokens to use this model so they can go
and test it before it's released in the public, which
seems like probably a very good idea, and they may

(04:51):
need more, who knows, but they'll be fascinating to see
what happens with this. Thank you, Paul. That is Paul Stenhouse,
our texpert this morning. Mythos is the name of that
new model from Anthropic. It has not been released to
the general public and it might be some time before
it is.

Speaker 1 (05:05):
For more from Saturday Morning with Jack Tame, listen live
to News Talks at B from nine am Saturday, or
follow the podcast on iHeartRadio