May 19, 2026 · 31 min · 12 segments
OpenAI unsealed the text messages. And the courtroom drama around Elon Musk, Sam Altman, and OpenAI somehow got even messier. In Episode 78 of What the AI?!, Jeff Keltner and Annie Delgado break…
Annie DelgadoHostUm, it's built on GPT 5.5 plus a custom kind of Codex harness, and they did say they have like a 5.5 cyber...
It's got to be like a 5.5 cyber and 5.5 cyber plus super-duper extra special version.
Um, but it's got a specific, uh, harness for Codex that's kind of a cyber defense threat analysis, and supposed to be really, really e- equally capable by many assessments to Mythos in terms of finding and helping patch security vulnerabilities.
Yeah, and a totally different kind of go-to-market, if you want to call it that, strategies between, um, Daybreak and Mythos as well, right? 'Cause we talked about Mythos last week and how, you know, they're kind of deploying it with the government, and we talked about that whole dichotomy.
They've named a bunch of launch partners, including Akamai, Cisco, Cloudflare, CrowdStrike, so a lot of the, you know, Palo Alto Networks, so a lot of the kind of large infrastructure and cybersecurity firms out there engaging with them on this.
Um, and so it'll be really interesting to see how people end up adopting using one, both.
I mean, I know a lot of coders who like, you know, plan in Claude and execute in Codex or vice versa, 'cause they kind of want the two models competing against each other.
So I'll be really curious to see what the cyber community comes up with in terms of how they use these models and what they think of the kind of Daybreak and Glasswing organization, whatever you want to call that kind of group of partners, and how they're working with the industry.
But it seems like they're both at least trying at the beginning to keep it on the defensive, fix the bugs, and try and keep it out of the bad guys' hands to, to use these things [chuckles] to, to be offensive and attacking.
Yeah, it's, it's really interesting because the, the tools, from what we can... from what we've heard at least, are extremely powerful in terms of finding vulnerabilities, and so that is the tension with like this OpenAI go-to-market strategy being more public.
Are you just sort of like giving the playbook, putting the playbook in, in bad guys' hands type of things?
Um, it's built on GPT 5.5 plus a custom kind of Codex harness, and they did say they have like a 5.5 cyber...
It's got to be like a 5.5 cyber and 5.5 cyber plus super-duper extra special version.
Um, but it's got a specific, uh, harness for Codex that's kind of a cyber defense threat analysis, and supposed to be really, really e- equally capable by many assessments to Mythos in terms of finding and helping patch security vulnerabilities.
Yeah, and a totally different kind of go-to-market, if you want to call it that, strategies between, um, Daybreak and Mythos as well, right? 'Cause we talked about Mythos last week and how, you know, they're kind of deploying it with the government, and we talked about that whole dichotomy.
They've named a bunch of launch partners, including Akamai, Cisco, Cloudflare, CrowdStrike, so a lot of the, you know, Palo Alto Networks, so a lot of the kind of large infrastructure and cybersecurity firms out there engaging with them on this.
Um, and so it'll be really interesting to see how people end up adopting using one, both.
I mean, I know a lot of coders who like, you know, plan in Claude and execute in Codex or vice versa, 'cause they kind of want the two models competing against each other.
So I'll be really curious to see what the cyber community comes up with in terms of how they use these models and what they think of the kind of Daybreak and Glasswing organization, whatever you want to call that kind of group of partners, and how they're working with the industry.
But it seems like they're both at least trying at the beginning to keep it on the defensive, fix the bugs, and try and keep it out of the bad guys' hands to, to use these things [chuckles] to, to be offensive and attacking.
Yeah, it's, it's really interesting because the, the tools, from what we can... from what we've heard at least, are extremely powerful in terms of finding vulnerabilities, and so that is the tension with like this OpenAI go-to-market strategy being more public.
Are you just sort of like giving the playbook, putting the playbook in, in bad guys' hands type of things?
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.