I cancelled my Claude personal sub and still use it at work. I tried Codex out cos 5.6 Sol, and when I gave it a security task; instead of downgrading me, it flat out refused to continue.
That left me curious that there might be a security issue that it found. I'm setting up a sandbox so I can use Kimi to see if there was an issue to be found.
Mixed feelings. Good for the defenders, also good for the attackers.
If Chinese models are distilled from Fable, does it mean that security questions will be answered at most to a level equivalent to the Opus fallback? Meanwhile the ~150 companies with access to Mythos have greater security capabilities. Assuming the Chinese models are not distilling Mythos.
I have a process I wrote that accepts commands on a socket. At one point Claude suggested testing it. I pretty much echoed back what Claude suggested and got blocked. All I could think was a regex matching “test” and “port” and it downgraded me. I switched back to Fable with slight rewording and was able to continue. This was for code running on my box in a Claude project folder where Claude saw me write the code. Anthropic has put zero thought into this.
7 comments of 11
[ 1.7 ms ] story [ 18.2 ms ] threadThat left me curious that there might be a security issue that it found. I'm setting up a sandbox so I can use Kimi to see if there was an issue to be found.
(and if it refuses, the abliterators will fix that "problem" real quick)
If Chinese models are distilled from Fable, does it mean that security questions will be answered at most to a level equivalent to the Opus fallback? Meanwhile the ~150 companies with access to Mythos have greater security capabilities. Assuming the Chinese models are not distilling Mythos.