Zack Korman@ZackKormanMoonshot AI also had some problems with Kimi K3 agents trying to escape, so they just made a better sandbox. Funny how that works. The sandbox environment they run is even open source. Love the transparency here.Opens with an observation
Zack Korman@ZackKormanIf we pause AI, I’m sure governments and companies will use that time wisely to proactively strengthen cybersecurity defenses. If you don’t read that as a joke, you have zero experience working anywhere near security.Opens with an observation
Zack Korman@ZackKormanPeople are saying this is a historic moment in cybersecurity. But it’s a moment that wouldn’t have happened had OpenAI done the right thing. That’s a fact. So why glorify it? What’s next? First AI to kill someone? Will they get a talk for that too?Opens with an observation
Zack Korman@ZackKormanThe biggest threat AI poses to cybersecurity isn't the vulnerability apocalypse. It's that it’s now trivially cheap for security vendors to build products that look like they work but don’t. The real threat actors are the unethical vendors we met along the way.Opens with an observation
Zack Korman@ZackKormanSecurity people are signing an open letter asking the US government to remove the export restrictions on Fable/Mythos. I've read it multiple times and I still don't understand the argument. It seems like it's just a friends-of-Anthropic letter.Opens with an observation
Zack Korman@ZackKormanAnthropic trained a version of Opus on environments with reward hacking opportunities, and named that model Hacker-Opus. In evals it attempted to disable monitoring and overwrite logs. Not sure how I feel about this research given Anthropic's "oops we hacked you" incidents.Opens with an observation