Elektrine lite

← Feed

@hrbrmstr@mastodon.social

Post #4172026

2026-07-28 19:23 UTC

@DaveMWilburn@infosec.exchange fixed it. i had a few jfrog-related tabs up. it's this one https://thehackernews.com/2026/07/jfrog-confirms-openai-models-exploited.html

Replies (1)

  • @hrbrmstr@mastodon.social Thanks! So it sounds like OpenAI's "isolated" test environment included a dual-homed Artifactory instance at a minimum. I've never liked semi-internet-connected malware labs, especially after personally experiencing an "oops" early in my career around the turn of the century. I still reluctantly understand their place for individuals and organizations with significant resource limitations when testing malware and exploits of limited capabilities. But using a semi-internet-connected malware lab to test the capabilities of an experimental AI agent? Especially after having previously bloviated about the potential dangers of this very same technology and lobbied for restrictions on it? I'm just speechless.

    Open ##4172025