Top
Best
New

Posted by Ryan5453 3 hours ago

Project Glasswing: Securing critical software for the AI era(www.anthropic.com)
338 points | 133 commentspage 2
Miraste 3 hours ago|
>We plan to launch new safeguards with an upcoming Claude Opus model, allowing us to improve and refine them with a model that does not pose the same level of risk as Mythos Preview2.

This seems like the real news. Are they saying they're going to release an intentionally degraded model as the next Opus? Big opportunity for the other labs, if that's true.

SheinhardtWigCo 59 minutes ago||
The other labs already censor their models. Everyone is trying to find the sweet spot where performance and ‘alignment’ are both maximized. This seems no different
wslh 1 hour ago|||
> Big opportunity for the other labs, if that's true.

It sounds like this is considered military grade technology as cryptography in the 90s. The big difference is it's very expensive to create, and run those models. It's not about the algorithm. If the story rhymes it could be a big opportunity to other regions in the world.

zb3 1 hour ago||
Well since Anthropic treats us as second class evil citizens, I guess they don't want our evil money either.
underdeserver 1 hour ago||
Interesting also is what they didn't find, e.g. a Linux network stack remote code execution vulnerability. I wonder if Mythos is good enough that there really isn't one.
zachperkel 3 hours ago||
Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser.

Scary but also cool

fsflover 2 hours ago||
Every piece of software definitely has serious vulnerabilities, perfection is not achievable. Fortunately we have another approach to security: security through compartmentalization. See: https://qubes-os.org
dakolli 2 hours ago||
Or more likely, its just an exaggeration or lie.
taupi 3 hours ago||
Part of me wonders if they're not releasing it for safety reasons, but just because it's too expensive to serve. Why not both?
coffeebeqn 1 hour ago|
If these numbers are correct it’s probably worth the extra price
Ryan5453 3 hours ago||
Pricing for Mythos Preview is $25/$125, so cheaper than GPT 4.5 ($75/$150) and GPT 5.4 Pro ($30/$180)
conradkay 2 hours ago||
For comparison, 5x the cost of Opus 4.6, and 1.67x for Opus 4.1

I think this would be very heavily used if they released it, completely unlike GPT 4.5

adi_kurian 1 hour ago||
Opus 4 & 4.1 are still on Vertex+Bedrock @ $75/1mm out. They were used very heavily and in my subjective opinion are better than 4.5 and 4.6.
breakingcups 30 minutes ago||
Interesting, what makes them better to you?
cassianoleal 3 hours ago||
Where did you get that from?

From TFA:

> We do not plan to make Claude Mythos Preview generally available

Tiberium 3 hours ago||
From the article:

> Anthropic’s commitment of $100M in model usage credits to Project Glasswing and additional participants will cover substantial usage throughout this research preview. Afterward, Claude Mythos Preview will be available to participants at $25/$125 per million input/output tokens (participants can access the model on the Claude API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry).

underdeserver 2 hours ago||
Key point: available to participants.
conradkay 2 hours ago||
permanent underclass has arrived :(
jFriedensreich 2 hours ago||
The only thing reassuring is the Apache and Linux foundation setups. Lets hope this is not just an appeasing mention but more fundamental. If there are really models too dangerous to release to the public, companies like oracle, amazon and microsoft would absolutely use this exclusive power to not just fix their holes but to damage their competitors.
zb3 1 hour ago||
BTW it seems they forgot about the part that defense uses of the model also need to be safeguarded from people. Because what if a bad person from a bad country tries to defend against peaceful attacks from a good country like the US? That would be a tragedy, so we need to limit defensive capabilities too.
Sateeshm 2 hours ago||
The bars have solid fill for Mythos and cross shaded for Opus 4.6. Makes the difference feel more than it actually is.
zb3 1 hour ago|
> On the global stage, state-sponsored attacks from actors like China, Iran, North Korea, and Russia have threatened to compromise the infrastructure that underpins both civilian life and military readiness.

Yeah, makes sense. Those countries are bad because they execute state-sponsored cyber attacks, the US and Israel on the other hand are good, they only execute state-sponsored defense.

More comments...