Teortaxes frames AI code capabilities as all-or-nothing: models either excel at spotting and patching vulnerabilities or they do not, turning apparent jailbreaks into straightforward prompts like 'fix this code' that expose the same weaknesses across frontier systems.
A very tasty update package from Zvi and yes… they're beginning to understand. This is the whole damn point of "superhuman hackers". To close the whole topic of hacking for good.
http://x.com/i/article/2067602345765060608
Prompts that coax models into revealing exploits function identically to ordinary debugging requests, erasing any practical line between offensive discovery and defensive repair.
Community comparisons note Fable locating identical code weaknesses to GPT-5.5 and Opus 4.8 variants, though post-launch export controls and access changes leave lasting availability unclear.
Users discuss treating AI jailbreaks as code fixes with some optimistic about formal verification enabling unbreakable security while others call related premises vile and criticize government caution for slowing progress.
No Digg Deeper questions have been answered for this story yet.
hey @TheZvi, if I may GLM is the strongest Chinese lab (at this specific moment) and this really is a frontier model. It is ≈Opus 4.7 in almost all text-only ways. Is reduces the gap more than R1 did at its time. Do pay attention, we don't want to repeat the same mistakes do we.
@TheZvi Glasswing isn't that btw. Glasswing exists to make sure the US isn't brought low by GLM 5.4 putting ransomware in every MRI scanner and iPhone. The current ecosystem is flimsy trash. For how we even begin getting to that, read https://ifp.org/the-great-refactor/
@TheZvi securing even all web-exposed software is a superhuman task, obviously. but the main issue is labor cost. The endgame is that the weakest link in the attacked system is post-quantum cryptography and you'd need to disprove contemporary mathematics to get anywhere.
@TheZvi This is indeed a very serious Q Yes, I believe that, because formal verification exists. This is a bit different from vulnerability search, but the proof of concept is a general purpose 10T LLM that tops MathArena anyway. Fable or Fable+ could write *unbreakable* software.
@teortaxesTex Serious Q: Do you think that a superhuman hacker could reliably write widely used and useful code so secure that not even that same hacker could break in even with 100x or more the effort that went into defending it? If so, why?
@teortaxesTex Serious Q: Do you think that a superhuman hacker could reliably write widely used and useful code so secure that not even that same hacker could break in even with 100x or more the effort that went into defending it? If so, why?
@teortaxesTex finding a numerical precision but and finding an set of overflows and races that make up an exploit is the same exact thing
A very tasty update package from Zvi and yes… they're beginning to understand. This is the whole damn point of "superhuman hackers". To close the whole topic of hacking for good.
@TheZvi securing even all web-exposed software is a superhuman task, obviously. but the main issue is labor cost. The endgame is that the weakest link in the attacked system is post-quantum cryptography and you'd need to disprove contemporary mathematics to get anywhere.
@teortaxesTex @TheZvi My argument is "echo" is not safe. It can emit "dangerous" code. We must ban the echo utility ...
There is a vile snuck premise that these LLMs have enough AGI to do things they were never designed to do.
@teortaxesTex Kinda makes sense why governments are so behind in tech if their response on lets fix something is “no, its too dangerous”
@teortaxesTex A meme must be made of this
@vxunderground
