Even if your country prevents access to compute to protect the trillion dollar companies, it’s not going to apply for every country, and as models get better it becomes easier to compete. There’s no way an AI non proliferation treaty will be passed or even enforceable.
What matters isn't the power of the tool, but whether defenders have had time to secure against. Today's cyberweapon is tomorrow's laughably obsolete.
Stuxnet used to be a national security threat, now I'm not sure it would be useful for anything.
State sponsored, non-public penetration fine tunes (of possibly public ones) likely can do it even faster.
Unsupervised penetration RL loop is ideal setup similar to optimization one – it's relatively easy to gain function on it.
And the fact that all our systems are riddled with security holes shouldn't be too much of a surprise given the way that we all know that software is developed and how tech debt / chores are constantly underbudgeted (plus I think this underscores that any one human's knowledge and attention are inherently limited, and even the best PR review is going to leak all kinds of security holes).
- With a weaker model, the time to break into the system might grow so larger that it becomes infeasible, similar to how password hashes can be bruteforced, but if the password is long enough, that is not going to happen in our lifetime.
- There might be problems which are inherently unsolvable with a lower level of intelligence. For example, your dog won't derive calculus from scratch, even if it lived forever.
- LLMs might be biased in such a way that they never explore the entire solution space, no matter how many attempts are made. Some models are notorious for getting stuck in a loop, trying small variations of the same approach every time, even though it is doomed to fail. This can be counteracted somewhat with higher sampling temperature, but that hurts reasoning capabilities.
https://www.csun.edu/~dgray/BE528/Pennigs2003Dogs_Calculus.p...
We're not talking about dogs, but LLM systems.
Mythos is not exploring entire solution space either.
Usually looping is solved by repetition/frequency/presence/n-gram penalties/DRY/min-p sampling, not temperature but we're not talking about small models that have those classes of issues here.
Let's just take GPT 5.5 and Opus 4.8 as an example. Both are worse than Mythos 5, but they're capable of quite a bit when the guardrails are lifted and they're paired with a skilled human operator. They more than "good enough" to reach the same result with the addition of some human effort.
> NSA director: 'Mythos "broke into almost all of our classified systems in hours"
> Donald Trump’s blocking of Anthropic is capricious and chaotic
So you either posted the wrong link or are just spreading FUD.
Those "tapes" DOGE took away? Nothing on them can be considered private any more. That's how brute force risk happens. Mythos' risks are showing doorways to exfiltration surely? Why bother when you can walk out the door with a data dump?
The NSA is just a highly specific subclass of the problem. Their traditional publicly stated approach to security is "nothing electronic which enters our domain leaves" and yet somehow they have assessed these systems as capable of breaching their walls? That's super bad.
I suspect they ran an analogue/instance inside their protection rings. I doubt they ran a test outside in the global internet. If they have actually lost control of their boundary, that's a bigger story (which I doubt) and contextually he could have been referring to information systems in NSAs duty of care, not things inside Ft Meade.
