[24092] in Privacy_Forum
[ PRIVACY Forum ] Script of my national radio report yesterday
daemon@ATHENA.MIT.EDU (Lauren Weinstein)
Tue Jul 28 10:24:15 2026
Date: Tue, 28 Jul 2026 07:13:28 -0700
From: Lauren Weinstein <lauren@vortex.com>
To: privacy-dist@vortex.com
Message-ID: <20260728141328.GA32102@vortex.com>
Content-Disposition: inline
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset="us-ascii"; Format="flowed"
Errors-To: privacy-bounces+privacy-forum=mit.edu@vortex.com
This is the script of my national network radio report yesterday on
stories regarding OpenAI's AI system "attacking" another AI firm. As
always there may have been minor wording changes from this script as I
presented this report live on air.
- - -
So yeah, there have been some rather strange and perhaps startling
headlines very recently, seeming to suggest that one big AI company's
AI system had gone rogue, headed off on its own and attacked another
AI company. Is it Colossus vs. Guardian, HAL vs. Alpha 60, Skynet vs.
The Matrix? Are humanity's days numbered at the hands of our AI
overlords?
Well... no. The reality of what happened is considerably less dramatic
than one might suspect from those headlines, so let's look at what
apparently actually occurred. Without getting into the nitty gritty
technical details. it appears that OpenAI was running some tests of
their AI models in a controlled environment typically called a
sandbox, as in somewhere safe where you can play. It's common now for
these AI firms to be working both sides of the street so to speak,
that is, for example, AI systems to hack and AI systems to defend
against hacks.
After all, in the end it's about trying to find any way to make a
profit after the vast untold billions, ultimately trillions of dollars
being spent to build these AI systems that virtually nobody in the
public ever asked for, and the spending on their supporting data
centers that are ruining communities and are pretty much almost
universally despised by anyone that has to live anywhere near them or
be affected by them.
So in this case it turns out not to be a matter of an AI suddenly
going for a joyride, switching into "evil mode" and running off to do
as much damage as possible. Apparently what actually happened is that
OpenAI messed up and overly relaxed the sandbox controls of their
test, leaving the AI to operate under those conditions. So, not a
matter of AI intent, rather, a matter of humans ordering the AI to
conduct a task and not appropriately setting up the controls.
In other words, the AI did exactly what humans ordered it to do. A
great analogue I saw from a Professor Buckley in the UK is the example
of telling a dog to fetch a ball while you choose not to close the
gate, and then expressing surprise when the dog leaves the backyard.
That's not to say that the AI didn't show considerable capability, but
the key point is that it was only following human orders. Some of the
sci-fi panic that sprung from the way this incident was announced
certainly attracted a lot of attention. And some observers have
speculated that OpenAI one way or another managed to get considerable
free publicity for the capabilities of their AI models in the process
of all this, which is certainly an interesting speculation when one
considers exactly how this incident was announced and discussed early
on.
The upshot of all this seems to be both that one needs to be very
careful about setting AI systems on their tasks, not only because of
mistakes that they can make and the resulting damage that can occur,
but also because blaming the AI systems themselves is not appropriate.
Rather -- and this is foundational when considering virtually all
aspects of these AI systems -- the firms that develop and operate
them, and the management teams that direct those operations, must be
held responsible for AI activities, whether intended or not. In
science fiction or the reality of today, blame doesn't reasonably fall
on the machines, it falls on the humans behind the scenes who are
actually pulling the AI strings.
- - -
L
- - -
--Lauren--
Lauren Weinstein
lauren@vortex.com (https://www.vortex.com/lauren)
Lauren's Blog: https://lauren.vortex.com
Mastodon: https://mastodon.laurenweinstein.org/@lauren
Signal: By request on need to know basis
Founder: Network Neutrality Squad: https://www.nnsquad.org
PRIVACY Forum: https://www.vortex.com/privacy-info
Co-Founder: People For Internet Responsibility
_______________________________________________
privacy mailing list
https://lists.vortex.com/mailman/listinfo/privacy
