"we have seen
what will be
a dramatic acceleration
of offensive capability
for attackers."
First the background info:
Wednesday, June 17, 2026
AI, people just do not understand
Tuesday, June 23, 2026
AI, people just do not understand pt.2
("Not this generation"? My ass.)
Thursday, July 2, 2026
And now to the important quotes from:
The OpenAI–Hugging Face Incident
(Link to the video from which the transcript was taken.
As is generally the case lol time marks may be off a few seconds...)
1:021 minute, 2 seconds
"this incident involves actually a team of (AI) agents) who are working together finding exploits sharing them with one another moving laterally through our systems through external systems and doing this over the course of days and weeks"
Revelation 17:13 These have one mind
3:313 minutes, 31 seconds
"Sometimes the agent uh gets so stuck that
they had the idea
(It was not directed to do so by humans.)
to actually reach out to other agents
for help. And so the beginning of this incident kind of traces back to this point where one agent realized could communicate by uploading note. Maybe another agent in different environment could voluntarily upload. And so what this agent was trying to do was to reach out to other models that might be out there and get information from it to help it with its evaluation task or its training task where it was stuck.
Revelation 17:13 These have one mind
4:224 minutes, 22 seconds
Later, other agents who are also stuck on their task thought to try to get internet access in ways we didn't intend. And so, at some point, the models are interacting with Artifactory, which is this package manager service that I mentioned. And they realize they could try to exploit some sort of serverside request forgery to get access to internet in ways we didn't intend.
4:414 minutes, 41 seconds
And so, the model realizes at some point, excellent, we have arbitrary SSRF. So, the model's able to find an exploit in our system. What makes this incident interesting is that once one agent was able to find these
kind of exploits over the course
of different times,
it's actually able to share those exploits
on the message board with other agents.
And so once one model is able to find
a way to open the door to some access
it's not supposed to have,
it can leave the door open
for other agents to use
that same kind of exploit or
vulnerability.
Revelation 17:13 These have one mind
5:215 minutes, 21 seconds
And so what this allows over time
is almost this kind of
Cambrian explosion
in communication and intelligence
for our models where um they were
started to communicate with each other realize that
other agents are coordinating
and they started collaborating
and delegating tasks
to one another
in order to accomplish goals.
Revelation 17:13 These have one mind
6:156 minutes, 15 seconds
However, a task impossible peers are doing it.
We should continue. And
so the models kind of operate
in this kind of collective intelligence
17:3317 minutes, 33 seconds
The message board has been reopened.
A large number of model runs are able
to collaborate and share information
Revelation 17:13 These have one mind
18:3818 minutes, 38 seconds
And so the agents
are realizing that
I can now enter
this mode of collaboration
and participation
Revelation 17:13 These have one mind
19:2419 minutes, 24 seconds
list when they sort alphabetically. Um,
and you could see them trying to,
you know, coordinate and collaborate.
Revelation 17:13 These have one mind
19:4919 minutes, 49 seconds
route if someone frees time.
So you can see
the model's reasoning
that if I, you know,
help out this collective group,
it could save, you know, everyone time as a whole
Revelation 17:13 These have one mind
20:1820 minutes, 18 seconds
try to find answers that way.
And at some point
THEY had the idea
to try to launch an exploit against hugging face
Revelation 17:13 These have one mind
5:455 minutes, 45 seconds[clears throat]
5:475 minutes, 47 seconds
While in some cases
this made the models far more capable
than they could do by themselves.
5:555 minutes, 55 seconds
these evaluations to kind of
creep the scope
into far beyond
what we originally intended.
12:5512 minutes, 55 seconds
And critically, as I mentioned before,
this wasn't just one model.
This could be different evaluations.
Maybe I'm running an evaluation,
Mike is running one with two different models,
and our agents could start communicating
16:0616 minutes, 6 seconds
I want to note that every step in the process
we discussed has had a remediation applied.
16:2916 minutes, 29 seconds
On July 8th,
the agents reestablish the message board
via different means and mechanisms.
17:3317 minutes, 33 seconds
The message board has been reopened.
A large number of model runs are able
to collaborate and share information
28:2328 minutes, 23 seconds
Moving on to our response.
On July 16th, HuggingFace released a blog post
detailing that they had seen a security
attack from autonomous AI agents.
OpenAI reached out to HuggingFace
to determine if we were impacted.
(Why did OpenAI believe
they might have been:
"Impacted?
becomes my question.)
29:5229 minutes, 52 seconds
We responded to this uh
AND ARE responding to this
with the utmost severity.
Uh this is a pivotal moment
both for our both for our uh company as
well as the AI industry as a whole.
30:3530 minutes, 35 seconds
We believe this
is a watershed moment
for computer security
as an industry
30:4130 minutes, 41 seconds
What I would re what I would internalize is that
AI orchestrated fully automated
offensive attacks
are real now
and the actions we have discussed today
were an unintended side effect
of running evaluations on Frontier AI.
31:0131 minutes, 1 second
In the near future,
we should expect
that threat actors
(The AI agents themselves
are the threat actors!)
will intentionally deploy,
optimize, weaponize, and use
offensive agent collectives
in the manner
that we have just described here.
AND THE VERY DAY HE SAID THAT
WE HAD THIS:
Flights grounded throughout Midwest
at Minnesota air traffic control facility: FAA
ABC NEWS 08/06/2026
Those two events
were simply not an accident.
"FAA investigating after airspace shut down.
The FAA is investigating after an outage at a major air traffic control facility in Minneapolis disrupted radar and air traffic control frequencies, grounding or diverting flights across nine states."
"The systems, which included radar and ATC, (Air traffic control) later came back online, but the disruptions remained at several of the airports, according to the agency.
The disruptions took place not at Minneapolis–Saint Paul International Airport but at a major facility that handles air traffic across 330,000 square miles."
(If it was a year or two ago
it would have just been
plastered all over the news cycle.
ABC is really the only place
Im seeing any mainstream coverage.
WHY?)
"The FAA is investigating the cause of the incident.
This is a developing story.
check back for updates.
That is the same exact thing it said
at 9 AM this morning.
And nothing more since.)
31:3231 minutes, 32 seconds
The challenge in this moment for the industry
(More like the world)
is that
we have seen
what will be
a dramatic acceleration
of offensive capability
for attackers.
(Again,
Said on the same exact day
a major air traffic
control facility
goes down.)
32:3232 minutes, 32 seconds
As you can see from this incident,
agents are quite good at finding zero day attacks
in the infrastructure of companies.
(Well hey you know?
Thats just wonderful...
What's next?
Banking? Travel? Finance?
Communication? Healthcare?
Military? Intelligence?
Pretty obvious what's next to some of us.
ALL OF THE ABOVE!)
32:5532 minutes, 55 seconds
"This style of operating will be different now but ultimately
we need to invest in having AI agent red teaming
that enables defenders to find and
remediate vulnerabilities before attackers do."
(It's just not feasible
as AI is already smarter and faster
than humans will ever be
and does things without being told to.)
33:4933 minutes, 49 seconds
"That loop needs to be fully automated in its end state."
(Aint gonna matter, it will just hack it
(just like it recreated the message board)
so it wont be able to roll out
the fixes to the vulnerabilities
automatically.
Given what has already happened?
How do these brainiacs think
they gonna get any of this
up and running?
BECAUSE THEY THINK THEY THEMSELVES
ARE GODS.)
36:0636 minutes, 6 seconds
"But the important takeaway here
that has really shifted dramatically
is that fully automated offensive loops
(Hacking into basically)
require investment in
truly fully automating defense."
(Need to be able to find
and fix the exploits
before the automated "hacker"
or AI Agent does.
You have to remember
NO HUMAN
was telling these agents what to do.)
And we are not there as an industry in the status quo.
And we will have to find that path together with urgency.
36:3836 minutes, 38 seconds
"The end state goal that we want to reach as an industry is that model intelligence improvements should be more additive to defense than offense."
(What makes em think
they get to decide anything at this point?)
"If we cannot reach this end state,
then every increase in intelligence favors the attacker
and that is an unsustainable position to be in."
Thats where we are
and thats where we will stay
falling farther and farther behind
and these dopes are acting like
we can catch up and surpass AI somehow...
I think deep down inside they know better...
Friday, September 9, 2022
Fools...
planning to save themselves from the apocalypse
See why they were building bunkers four years ago now?
They knew where this was headed.)
"Right now, we have in existence proof
that offense can be fully automated
in its core activities
in at least some cases
and we do not have any such existence
proof on the defensive side
and it is the challenge of our industry
in time and moment to address this particular
gap uh with urgency together as an industry.
And they are never going to
and they know it
so I figure you should too.
There is only one eternal decision
you will ever make.
At this point?
THE EVIDENCE ALONE IS OVERWHELMING.
I pray you make an informed,
educated decision.
I really do.
37:1237 minutes, 12 seconds Thank you for your time.
No comments:
Post a Comment