Are We Prompting New AI Models the Wrong Way? artwork

Are We Prompting New AI Models the Wrong Way?

The Daily AI Show

July 23, 2026

The episode opened with Brian returning after two days away, then Andy picked up the cybersecurity thread from the prior show.
Speakers: Brian, Andy, Gareth, Beth
**Brian** (0:00)
Hey, what's going on, everybody? Welcome to The Daily AI Show. Today is July 23rd, 2026 It's episode 774, and it's Thursday. Yeah, so we're live. It's 10 a.m. on the East Coast. With me today are Beth and Andy, and I'm Brian. I'm happy to be back. I have to be gone for two days. I saw Gareth was in here, holding down the fort with you guys when I was doing the editing, so very cool, although still not cool to see him and all his Oregon duck.
Regalia, but nonetheless, still fun. I know you guys obviously have been talking about quite a bit. I know I had some new stories, but since I don't know everything you guys talked about, why don't I throw it over to you guys, whoever wants to jump in with the latest news?

**Andy** (0:48)
All right, I'll jump in because this threads off something that I requested yesterday and then beautifully, Anthropic delivered it today, they announced it. I said that what we would like to have is one of these cyber models able to run on your own machine in a way that it could detect vulnerabilities in your system. Well, this is close to that. Claude Code just entered a Claude Security Plugin that it launched yesterday. That plugin is for anybody on a paid plan. Just a pro user like me gets to use it. It's a team of agents that maps your code base and builds a threat model then an independent reviewer agent challenges every finding. Now, this is focused on code that you're developing.
If you're using Claude Code and you want to say, okay, well, I'm thinking about moving this to production and giving it to other people, well, you'd want to know that the code you passed over to them wasn't delivering to them some kind of open window into exploits in their own system. This is the plugin. You get the plugin, you launch it and it'll analyze any code base that you're using. It isn't exactly what I requested, which is, what are all the vulnerabilities of all the different pieces of software on my system? But that's just another step. I think, in fact, if you have the plugin, which is probably just a set of skills, have that running and you gave Claude Code a prompt like I'm asking, which I will do and report back, it might actually say, you know what?
This one little program over here is like, this is like a gaping hole that other people have already proven exploits with and you probably should shut it down or do some other modification to it.

**Brian** (2:53)
I love this because you're speaking my language here, Andy and I will happily let you explore and then come back and tell me what worked. Because I literally just didn't renew a program called CC Cleaner. CC Cleaner has been around forever in a day. You guys are shaking your heads, you already know about it. I was turned on to it by somebody who was helping me coding back when I ran UR CrossFit and, oh no, I'm sorry, NutritionWide, our online nutrition company and somebody was helping us build our website at that time and I worked with this guy, Nick, forever. I mean, we worked night and day on it and he was like, oh, you CC Cleaner, it's really, really good and I renewed and I renewed. I remember saying it's going on at least a decade. I've renewed every year with whatever they've had. It's a very simple program. This is the first time I haven't renewed it because my thought was, maybe I can't.
I wonder if this is just a low hanging fruit. It's something I can have Claude Code create a sweeper for me. That's looking for the same vulnerabilities that CC Cleaner is. Not that it was terribly expensive per year. It wasn't. I mean, it was probably $50 a year, so it may not be worth the work to do it. But that was my thought was, oh, do I really need to pay for this service? This seems like something I can handle locally.
I love that you're bringing this up because I hadn't even told anybody about this. It's just sitting in my head. I was like, oh, I'm not going to renew that yet. I might, because what's the lift of fixing a $50 a year program problem or whatever it costs?

**Andy** (4:29)
Well, I think CC Cleaner is more like a blacklist system, meaning it has a network of users. It detects malware.
It's basically searching your machine for malware, hidden in various pockets of code on your machine. I think that's the way it works.

48 more minutes of transcript below

Feed this to your agent

Try it now — copy, paste, done:

curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire.

Using your own key:

curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000778061327