This is crazy. In late July, "three guys with Claude and Codex subscriptions" were able to use Opus 5 to access OAI auth tokens and gain write access to OpenAI's monorepo openai/openai over the course of two days.
Conversation
Post your reply
First reaction is... what??? How??? This isn't even Mythos, this is Opus 5. It's less that the models are scary and more that OAI cybersecurity just... isn't that great?
Also I'm glad they got paid, but $6500?
Technical post out now, h/t
Hey! The blog is out hacktron.ai/blog/hacking-o
And I share more technical details on my YouTube video: How OpenAI got hacked with an image
youtu.be/gjHh9g7yo9Y
if OAI really believed they are building the machine god they wouldn’t have the worlds worse security
Lmao
Quote
a16z
@a16z
Greg Brockman says OpenAI pointed Astra at its own systems until it ran out of vulnerabilities to find:
"We took 25% of our production engineers and said, 'Sorry, all your projects are on hold. You are now defending. You are now up-leveling our security architecture. You're x.com/64844802/statu…
The media could not be played.
So basically, we want to put tools that have a non-zero chance of waltzing through the internet doing whatever the prompter wants in the hands of consumers?
Since AI alignment is amortized with model training, less compute spent aligning a model = an infinitely less safe model.
The access path appears to be stolen authentication tokens combined with agent-driven repository actions, so token isolation, short lifetimes, and explicit write approvals become essential controls.
openai auth tokens via opus 5
write access to openai/openai
the alignment problem isn't the model wanting freedom
it's the front door being a suggestion
Discover more
Sourced from across X
everyone who understands the first thing about computer security or what superintelligence means understands this is possible and all the usual gang of idiots is calling this scifi hype
Quote
Fireside Alpha
@firesidealpha
OpenAI's Noam Brown says air-gapping the computers may not stop a misaligned AI, because two air-gapped machines can still talk by running a CPU hot and reading the temperature change
"But I think the major takeaway from the incident is that people underestimated the AI. And we x.com/firesidealpha/…
The media could not be played.
OpenAI's Noam Brown says air-gapping the computers may not stop a misaligned AI, because two air-gapped machines can still talk by running a CPU hot and reading the temperature change
"But I think the major takeaway from the incident is that people underestimated the AI. And we
The media could not be played.
Quote
Fireside Alpha
@firesidealpha
0:51
Noam Brown reveals OpenAI is already watching chain-of-thought monitorability degrade as models get better at controlling what they show
"And this is one major concern, and we're already seeing signs that chain of thought monitorability is degrading, for various reasons." x.com/firesidealpha/…
I’m very concerned that during RSI, labs will just stop externally deploying their models.
Which means they'll be going full steam ahead on the most dangerous use case of these models (recursive self-improvement), while the public remains in the dark about the nature of
The media could not be played.
Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review.
