News for hackers. Stuff that matters.
Re:Qwen4-Max by drnick1 on Monday November 16, 2026 @01:53AM (#50514793)
Are these Chinese models still accepting arbitrary cybersecurity and reverse engineering tasks? I am increasingly frustrated with Claude's safeguards (and Codex, to a much lesser degree). It is flat out refusing to look at game binaries and anti-cheats for the sake of learning and devising Linux workarounds.
Re:Qwen4-Max by nolok on Monday November 16, 2026 @02:42AM (#50515082)
If they don't, they all have third party made abliterated versions available
Re:Qwen4-Max by ctolsen on Monday November 16, 2026 @04:35AM (#50515721)
Hosted where? In my experience they might put more guards when hosted by the company that makes them but if you use them through inference providers they do whatever you want.

Although, can’t say I’ve tried too hard (not keen on sending too much data to China) but I’ve been using K3.1 through Moonshot and I have had zero issues.

Mistral Large 4 (through Mistral) also doesn’t refuse a thing.

Re:Qwen4-Max by Havoc on Monday November 16, 2026 @09:43AM (#50518113)
Latest kimi had cyber nerfed intentionally

Not blocked just less training

Re:Qwen4-Max by Hackbraten on Monday November 16, 2026 @10:05AM (#50518370)
Anecdotal but qwen3.8-max has been grinding away at any RE task I’ve thrown at it since I switched from Claude three months ago.

(May not be representative because I’m extra cautious to frame all my RE requests explicitly as interop issues. Which they are, more or less.)

Re:Qwen4-Max by mewse-hn on Monday November 16, 2026 @12:27PM (#50520184)
> Are these (hosted) Chinese models still accepting arbitrary cybersecurity and reverse engineering tasks?

In my experience, yes. I wanted to flash my new wifi access point with openwrt and opus 5.5 refused to explore how to get root, so I switched to mimo-v2.6-pro and it figured it out.