Post caption
Anthropic has finally confirmed the existence of Claude Mythos, but they will not release the model publicly, because it isa “cybersecurity reckoning.”��Instead, they have organized a 40-company coalition called “Project Glassing” to allow cybersecurity professionals a head start in locking down critical software.
This is an urgent initiative and it’s time to pay attention to alignment, because these models are getting seriously capable.
Mythos’ benchmarks are absurd:
It scored 77.8% on SWE_bench Pro, versus 53.4% for Opus 4.6.
59% on SWE-bench multimodal, versus 27.1% for Opus 4.6
That should give you an idea.
It is a tremendous leap.
Imagine what you will be able to build with Mythos when it eventually drops.
Also imagine, the damage you could do with it If they were to prematurely release it…
#anthropic #claude #mythos
Transcript
Five hours ago, Anthropic announced that they built a model so good that it is quite literally powerful enough to break the Internet. It is the first time that an AI lab has refused to release a model because it is too dangerous. Not in a physical way, but but for cybersecurity. It is so capable that it found bugs in, quote, every major operating system and web browser, some that, for decades, no human security researcher ever found. Obviously, capabilities in a model like this could do harm if in the wrong hands. And no, we won't be releasing this model widely. Yeah. The cyber threat is so severe that they're scared to release the model publicly. So what are they doing instead? Well, Anthropic has started a 40 company coalition called Project Glasswing, including Apple, Amazon, and Microsoft to give them a head start in locking down critical software. More powerful models, no give them a head start in locking down critical software. More powerful models are gonna come from us and from others, and no we do need a plan to to to respond to this. And they added that mythos is only the beginning, that it is a starting point for a quote unquote reckoning. Rank what we're gonna see is an arms race between hackers and the companies racing to defend their system. This is an industry change point because, as they say, this is the worst AI will ever be.
