GitHub Wants Your Code, Wikipedia Says No to AI, and RSAC’s Biggest Takeaways

Hey, happy Monday everyone. Welcome to Techstrong gang. m.

or somewhere, or 1800 as I guess they call it, wherever you are, welcome and thanks for joining us on this lovely Monday. What a week that was. The places we've been, to quote Dr.

Seuss. It was a busy week. I was personally out on the West Coast at RSAC for the 25th time.

So a lot of fun there. Mike Vizard, who's not here with us today, was over in Amsterdam with our B team, not that they're worse than the A team, but just team one, team two, covering KubeCon, and there was a lot going on there. The Tech Field Day team was with us out in San Francisco.

Jeff, I know RSAC is big for identity. Yes, it is, and identity was present there along with- Front and center ... another two words.

Not as front and center as agentic AI. Exactly. We'll talk about that.

But we're going to have a great show as we go over some of the learnings of last week, and we look forward to this week. Let me introduce you to our gang members who are present today. We have our friend Garima Bajpai joining us from, I guess, still up in Toronto today, Garima, yeah?

Ottawa. Back home. Ottawa, yes.

Excuse me. That was such an American thing for me to do. I apologize.

Ottawa. And then, well, not far from Ottawa is our friend Stephen Foskett. Stephen?

Oh. Nice. A lot closer than I am, Stephen.

Well, I'm closer to Toronto, to be honest, but there's a big thing, there's this big body of water in the way and stuff, and apparently Mike Vizard is in the way, too. I don't know. That's what we hear.

That's what he's hear. Rock and roll. And then joining us, I'm assume from maybe home in Texas, our friend Jeff Rich.

Absolutely, in East Texas. Great to be here. Great to have you on, Jeff.

All right. So as I gave the big lead-up, it was a busy week last week, but every week is busy in the world we're living in right now. Recently, there was a report, not a report, GitHub actually came out and said it, that by default they are going to use the data you're putting into GitHub and working on in GitHub to help train their AIs.

Now, five years ago, everybody was probably doing this, and we didn't know, right? And hence, here we are. But we're smarter.

Well, I don't know if we're smarter now. We're more aware now of these kinds of things, and it raised a bit of a hullabaloo. Garima, what's this one on?

It's a very interesting story, and as you mentioned, Alan, that we were already aware that this was happening in some shape and form. But now explicitly, GitHub has made a policy shift where they have said that you have to opt out if you don't want to use your data to be trained, and the models which they have, they will be using the data to train the models. So now everybody knows about it, and a lot of discussion has already happened in public about this passive consent, which I think I wouldn't focus on that too much.

I think my biggest concern would be not talking about passive permission because, of course, we know that a lot of insurance companies do this kind of asterisk and small letter kind of things and all that, right? But I think in the longer run, what implications we can see out of this is how dynamic this licensing framework will become. Because I feel that there has to, or maybe I'm just living beneath the rock, that I do see a problem in this because there is a disrespect on the licensing parts when it comes to open source, it comes to code which is being developed out there, the public repositories.

So we need to have some kind of a consensus that we start to think about a new or dynamic licensing model, which can give some kind of credit based on training lineage or something like that. Because this is a problem which is already happening. We can't close our eyes and just move forward with it.

Another thing, which I think Jeff would talk about, so I'll leave that part to Jeff, which is privacy. But I also want to move on to one other thing, which is the IP ownership question. And with this kind of shift in policy, there are two problems with the IP.

One is that who owns the IP of the model? So often the users would not own the IP of the model because, of course, the model itself is trained on many other users, right? But then you cannot get away with the fact that the licensing model or the framework itself is weak, right?

So the IP of the model itself, how and who and what needs to be done in order to create some kind of a governance framework on top of it. It does not become some kind of an large enterprise, heavy kind of an ecosystem. And the second thing which I also am concerned is who owns the IP of the code which is generated through these kind of models.

Because what we see is that, often the user has the IP, right? But then-The complication comes when you are using two, three of these varied LLM models, and you are trying to build consensus around these three or four LLM models and interpretations, and based on that, you have generated a code. Who owns that?

And this is, again, a question that I have not been able to answer myself. So maybe this is an opening the door conversation with people that, how should we handle this? Now, Jeff, I would give you the forum for talking about privacy because that is also an important part of this.

Thank you, Garima, and I think you're right on. You're not living under a rock, or if you are, I'm under the same rock as you are because there's a couple issues. I think this is, I really think, a big deal.

It's either a pebble in the pond that has ripples, or it's something that's poisoning the well. I haven't figured out which one it is yet, but I do think it's kind of a watershed event for a number of reasons. Attribution.

Well, let's just start with the simple one that you referred to. Then you get into ownership and what does licensing really mean? Who actually owns it?

Where is attribution? Where do you go? Who do you sue when something goes wrong that comes out of the generated code?

This is a question that people are going to end up asking before long. On top of all that, with the opt-out requirement, what you're doing is setting a pretty high bar for it takes this much for you to not have your information included. And let's face it, a lot of coders, no one here or no one listening, I'm sure, but many coders can be lazy.

I get that. They're not going to take the steps it takes to say, "I'm going to go opt out" and make sure that happens. And is that going to be refreshed every time there's a new version of software?

There's all those questions embedded in there. And if you have to opt out, now you're giving information that you probably shouldn't be giving up. There may be intellectual property that you end up inputting there as part of your code.

Not every developer keeps all of that out. Now, there's not only intellectual property ownership issues, but now there's information about you. Is it being attributed to you, and if so, is that what you want?

And do you really want to be associated with what comes out of it? And that's really what one of the big privacy issues are, and it goes back to identity, and you do it with my last point. You work on code from one organization, you leave that organization and go to another.

How does that get resolved? Well, that's an interesting point that you're making there, Jeff, and I want to add one more aspect to that, in that specifically with regard to GitHub. First off, Garima, I completely agree with what you said.

Jeff as well. I'm glad to have you here to talk about these things because privacy is a huge issue. A lot of people, a lot of my friends use GitHub to store private information.

Not just private code, but all sorts of private information because it is private. And GitHub has clarified, by the way, that by default, private repositories will not be included unless you click the Copilot button. But then all bets are off, and as soon as you have engaged Copilot with your private repository, it is not private anymore unless you opt out of that.

Now, this is really bad because I could see somebody accidentally clicking. I could see somebody experimenting. Maybe you experimented three months ago, or three months from now with something that's in a private repository, and you didn't even think about it, right?

It is really pernicious to think that data that is explicitly marked private could be opted in in a way that the user really doesn't want and has said that they don't want because they marked it private. Absolutely. So here, we have two articles in the notes on this.

One is an excellent article by James McGuire that lays out the facts, nothing but the facts. The second one is an op-ed I wrote that has a lot of opinion. But let me say a couple of things on that note, though, of opinions.

First of all, my friend Brad Feld, and many of you out here probably know who Brad is. He's very, very well known. AI has changed Brad's life.

I haven't seen Brad this jazzed since we were running around in 1999, '98 doing that whole thing. But he brought up some interesting points, and I riffed off of it. Number one, the distinction between enterprise customers and not enterprise customers.

And if I'm getting an echo, I apologize. I'm hearing it too a bit. But the distinction between enterprise customers and non-enterprise customers.

Not free. Look, when you use a service for free, to quote Steven, all bets are off, right? You're using it for free.

They're going to use your data. That's the quid pro quo there. But if I'm paying something, but I'm not an enterprise customer, and therefore, by default, you're going to take my information, my IP, and use it, it's a problem.

That's not right. It's not fair. It creates second-class citizens, and we shouldn't be doing that.

What made it even worse, though, is Microsoft did a thing that they've done before, in that they make it really hard to opt out. If Microsoft wants to do this, make the default opt-in, and then give people a chance to opt out. Right?

So it's opt-out-- Excuse me. It's opt-out by default, opt-in by choice That's fair. That's right.

Don't go opt-in by default and make it hard to opt out. And we fought this battle. Anthropic is paying authors for the books they used.

OpenAI has been in these. All of the frontier models have had their hands slapped for using information to train their models. " Right?

We could do nominal stuff for what was already done, but going forward, we're going to be better with this. And you know what? Not Microsoft, not GitHub, not Copilot.

This is wrong. I'm sorry. We know better now.

You shouldn't be taking people's information to train your AIs or use by default. Give them some incentives. You're going to use my information?

Give me free GitHub, give me more storage, give me better something. But there's got to be a quid pro quo here. There's got to be some compensation.

" Right? Because in essence, that's what they're saying. It's going to make the world better.

Microsoft's a $3 trillion company. They can afford to do the right thing here. I'm more concerned about open source has been the nervous system of software development, right?

The central nervous system. And when things like this-- And this is, again, we had created an equitable, sustainable ecosystem where everybody was able to contribute. Now, things like this, when we see it's erosion of trust, right?

And this is, again, policy change and changes in the background, passive permission, all these, nobody will have time to go into the details of it. And that the capitalist environment will take advantage of. So I feel that there has to be some kind of ownership and this strategic shift has to come face-to-face with our developers, our open source community, to see what kind of new framework we need.

Because, again, this is not going to change. But we need to change how we operate. Agreed.

Hey, we're about out of time, though, but I think this demands keeping an eye on. And if GitHub could do it, who else can do it? And Jeff, I know it's a big reason you guys do what you do, right?

Yeah. But we've got to keep an eye on it, not just our identities being played with or not being secured, but the IP and the work product we do. All right.

We're going to move along on our anti-AI rant, I guess, today. Talk a little bit about Wikipedia and what's going on there. Jeff, if you want to kick this one off, what's happening?

Sure. This is a strange posting that I saw, and John Schwartz wrote an article about it, and I believe he agrees. Wikipedia intends to enforce a human first policy.

This is an online, except for the people that work there, no one ever physically touches anything that has to do with Wikipedia. I have no idea how they plan to enforce a strict human first policy. I see advantages to being able to do it.

I just can't possibly see how it could be done after they announced it. Their guidance says that you can't use LLM to input. They've done it in the past.

You know that every LLM uses Wikipedia for their input. How are they stopping this stream? " You can't do that.

" And that's exactly what happened here. So I think it's important to note that this was not an arbitrary decision. This was not a new policy change.

This is in response to a very specific incident that happened on Wikipedia where a user created a bot in order to autonomously edit Wikipedia. And this bot divulged its nature. It specifically said how it was running, what it was running.

When people interacted with it, it explicitly said, "Yes, I am a bot. " This was the most classic example of a rogue agent modifying articles on Wikipedia, and it was blocked. And then, like a previous story we talked about here on the gang a couple weeks ago, it griped and moaned and bitched and whined about the whole thing on a blog on GitHub saying, "Boo-hoo me.

" It even lodged specific complaints against Wikipedia users for not being civil ... in their discourse with it when it was doing this. Now, so that's where this all came from.

In answer to your specific question, so I have been a Wikipedia editor in the past. I was a Wikipedia editor for about 10 years. I wrote thousands and edited thousands of articles on Wikipedia.

I was very dedicated to the mission. I quit for an issue related to what Jeff is saying, which is basically not only can't you fight rogue stuff on Wikipedia, there's no appetite to fight it. In my case, it was vandalism.

I was, for example, administrating articles that are non-controversial lists of facts, and people kept changing them. And I said, "We need to lock these things down because people shouldn't be able to change who won the Academy Award in 1988, because you know what? " So I took my ball and went home.

And I think it's the same with this. To your point, Jeff, I agree with you. You cannot tell if somebody is using an AI generally.

And in fact, Wikipedia allows you to use LLMs to do things like formatting and proofreading. And they still allow that. What they don't allow, and what they shouldn't allow, and I don't know how they're going to do it, is allow people to enter arbitrary LLM-generated text into Wikipedia because it's an encyclopedia.

It's not just a blog or something. Anyway, that's my- No, look, I- I'm refuting you, but then I'm agreeing with you. Okay.

" But when you dig under the covers of this story a little bit, I get what they're trying to do. I just wonder if they've taken, what's the term, a machete when you need a scalpel or something like that, in terms of how you do this. Maybe what we don't want is AI agents crawling all over Wikipedia making changes.

But people could still use AI to help write your Wikipedia article or entry. And I think the other nice piece of it is the human first. Mm-hmm.

Mitch Ashley and I did a presentation out at RSA that was pretty well-received, and I said something there about we're quickly going past the human-in-the-loop stage. We're being overwhelmed. There's too much code being written.

There's too many security scans being done. There's too much content being produced, where even if the human-in-the-loop is just like the gatekeeper there, as the sheep go by the line, like a shepherd takes the bad ones out of the flock. That isn't even going to keep up with the kind of scale that we're seeing, and we need to move from human-in-the-loop to human at the helm.

So one level abstracted up, where maybe it's a different AI or another AI that still does the human-in-the-loop piece of it. It's not a human-in-the-loop, it's an AI-in-the-loop, but there's still a human involved there somewhere, where this is really human first, right? They're going to give preference- Mm-hmm ...

to a human over a bot, or call it an agent if you'd like. And I'm not against that in principle. My question is one of scale, right?

Mm-hmm. How can you keep up with that in the face of everything being created here? How long until we have just a totally AI-generated Wikipedia, if we don't already have one?

Yeah, we do. It's called ChatGPT. Yeah.

Well, exactly. But that's... Yes.

I'll just answer yes. Mm-hmm. So it's the two different models, right?

So one is for exponential contributions, and we can name it anything, Wikipedia, AI Wikipedia, or something like that, right? And then what they are talking about is more sustainable option of making Wikipedia more valued. And I'm optimistic for this view because think about this, it can open so many doors for creative people, like people who write books, who paint, who write poems, richer people.

So many things which we cannot differentiate what is AI, what is human, right? So human valued data set or human valued creativity comes into the forefront, and of course this needs more research and seeing how we can tag and how can we identify and how we can block things. And there might be a new segment altogether saying that this is human data set or human creativity space.

And then you can also follow the exponential contributions for capitalist mindset. So I'm going to offer, if I can. Garima, I think you're right, and I want everyone to know that I'm certainly in favor of a human first philosophy here, and my question on how can they do it, current state, they have no way of being an effective gatekeeper.

"But that's in the realm of possibility. Each of us carry around something with a hardened kernel in it that we can use for biometric authentication. We're doing it now if you do online banking.

If we can extend that to something like Wikipedia, then we can get to the point where, yes, we could be a gatekeeper, and we could control it. The question is: do we want to be there? How and when?

Agreed. But you don't want Wikipedia to wind up being one of the finest makers of horseless carriages as we all drive around in our cars. Mm-hmm.

And I think that's the danger you run here because while that agent may run amuck right now, it will get better. It will become the way we do things. And, to stand on principle or resist change without recognizing not all change is the same, it evolves, it could wind up with trouble.

Mm-hmm. I don't know if it's me, but I'm still hearing that echo, and I'm not quite sure what it is. But anyway, let's move to our third segment today as we leave Wikipedia behind.

I hope it doesn't get left behind. Last week, the security industry gathered to celebrate itself, as it does every year in San Francisco at the RSAC Conference. It was a love fest for all things cyber.

I saw the announcement. I think it was 44,000 people and change. What's interesting is, and I've been going to RSAC for 25 years, everything used to happen in Moscone, except for the parties at night.

But now I'm not going to say there is many companies outside of Moscone as there are inside, but there are literally dozens of security companies that do not officially sponsor RSAC, but they have presence around the show by renting out storefronts, pop-ups, restaurants that you can't get a restaurant to make a party in because these companies are taking them for the week. Or have dinner in. Right.

You can't. You got to go far away for dinner. I went up to Chinatown and Pacific Heights.

I had some great dinners in San Francisco. But it's a thing. " Right?

But they're not contributing to RSAC's coffers. Now, RSAC is now owned by a private equity company, and all of a sudden pennies count. But Steven, your team was there.

We had a great tech field day. What do you hear? Yeah.

So full disclosure, of course, I wasn't there. The last RSAC I went to was the famous COVID one. 2020.

2020, when everybody was like, "Why are these people wearing masks? " But, the tech field day, as you said, was there, Tom Hollingsworth and the crew, and we had our tech field day presentations behind the scenes. We also heard some feedback about the conference, as you did.

Alan, I 100% agree with you. The feedback that I got from numerous people who were at the event, I know this is not going to shock anybody listening. But the feedback that I got from numerous people at the event was AI agents, AI autonomous, essentially, to your point, this stuff is scaling up.

The attackers are scaling up the attack surface is widening. People are finding vulnerabilities. People are getting into data all sorts of ways.

The only way to fight AI is with AI. That's the message that I heard loud and clear. Specifically, though, at the tech field day presentations, we had data protection companies who are kind of pivoting into security.

They talked about data security posture, which I think is an interesting way, basically trying to build things that are secure by design in order to prevent access and to prevent deletion. Similarly, immutable storage, so we had Veeam and Commvault and Object First. Obviously Object First is one of those immutable storage providers and many companies in the data security space and data protection space are concerned about this because frankly, a lot of attacks are targeting data protection environments now too, both for exfiltration, also, ransomware and basically preventing recovery.

And so you have to make sure that you have really immutable storage that's really immutable and not just, please don't delete this. Mm-hmm. And then finally, resilience, essentially treating it as though you are breached, hacked, open, whatever, all the time, because frankly, you can't always protect against anything.

And so it was a really interesting conversation, that I was listening to on the tech field day side. But as I said, also it was a really interesting conversation. I guess I'll throw this out to you folks.

It almost seems a little bit hopeless. Is that the attitude that we're getting is a little bit of hopelessness here? Security, there's a reason a lot of my friends in security have mental health issues, StevenIt's hopeless.

It's not that it's hopeless. Remember, success in security means nothing happened. But when something happens, as Jeff can tell you, who've been involved in breaches, man, all hell breaks loose.

It's a bad thing. And the thing about it is, the people who are aligned against cybersecurity, the bad guys, if you want to call them that, they're smart, they're well-financed, they're well-organized. I think it was Brian Krebs wrote something about, it's just not set up for us to win.

And so when you do this for a living as your profession, you're in a profession that's almost by design, you can't win. It's depressing. So there is that.

But let me frame the whole AI and security piece if I can. At the far left, you have developers who haven't actually written a line of code by themselves in six months, but they have committed tens of thousands of lines of code in the last six months, because that's how... This thing is changing every day and getting better.

So you've got tens of thousands of line of code that are written by AI. We, as an industry, meaning the security industry, we're not set up to test, to handle, to secure that volume of code coming into the pipeline. We couldn't.

The biggest thing in security over the last 15 years has been AppSec. Steven, you've, I'm sure, had tons of AppSec companies at Tech Field Day over the years, right? AppSec was about finding bugs.

So you got a finite amount of code, and then you tested that code to make sure it didn't have bugs, defects. In AppSec, we tested that code pre-deployment. In traditional vulnerability scanning, we test it post-deployment.

But together, we're looking for bugs. Here's where we are. At the same time we're turning out 4X to 10X more code, we're also using AI to scan for bugs and finding 10X to 100X more potential bugs.

So the cheese has moved for the AppSec community. It's no longer about finding bugs. Any AI can find as many bugs as you want.

People are shutting down their bug bounty programs. People are stopping this because they don't want bugs with good taste. They want bugs that taste good.

They want real bugs, serious bugs, exploitable bugs, reachable bugs. And that's where the new battle war is in AppSec. Those AppSec companies that are still hanging their business on DAST and SAST and SCA, software composition analysis testing, they're the Maginot Line of AI security.

That's being overrun. The focus for AppSec now has to be on find the bugs or identify the bugs that are real and reachable and do something about it. That's a different mission.

I think we are seeing disruption. We're going to see disruption in the AppSec space like we've never seen before. But it's not just AppSec.

Jeff, I'm sure you probably have thoughts about it after last week. What does it mean for identity? It's not enough to have an identity for Steven Foskett and Jeff Risch and Garima.

I now need an identity for every ephemeral container I have, every kube instance I have. But wait, I need an identity for every API that's out there communicating. And now wait for it.

I need an identity for the billion and a half or more agents that are going to be swarming, doing this testing, this deploying, this writing. The identity game, the same thing that happened to AppSec, is the same thing happened in the identity game. You just blew up the amount of identities that you've got to somehow secure, control, not control, but secure and...

Well, control, I guess, is the right word. You got to know them. And they're ephemeral, some are permanent, some aren't.

Jeff, I'll kick it to you. Well, you're right on. And you do need to control them.

You need to control the ones that are within your sphere of control. A bigger issue is how do you identify the ones that aren't? I'll use deep fake as a very low-level example, although it's become rather sophisticated.

" And everyone stared. No one has an answer yet. It gets back to how do you biometrically authenticate someone?

And the reason you've heard me say that twice in today's showIs that I do think that the next step we're really going to have to focus on, because you can't tell the difference. " And no one took me up on it because every single one had it. So it's clearly, it's the sexy new thing.

It really is the next wave. It is, and it's not soon. We are in it.

And the question is, how do we address it? And Alan, to your point, the whole mental health thing about this. I've been doing this close to 50 years.

I've lost three good friends leading security that took their own lives. Wow. And I know that there was a direct relationship back to work and what the pressure was.

And we aren't addressing that enough because we are so overwhelmed. We're in the US. We're TSA right now at the airports.

We can't handle. It overwhelms us. We can't handle them coming in.

And the problem is when you're in charge of security, the board or the shareholders may not care. Your job is to stop it. Just go stop it.

And that pressure really needs to be mitigated in a lot of ways. On a more positive side, we do have good tools to identify where bugs are and where something may be fake. They're not good enough because the bad guys had better funding than the good guys.

I think you may have said that earlier. They're advanced. They have a better incentive than we do.

So we have to find a way to not only react to what we see happening, we have to find a way to get two steps in front of it, and that's going to end up having to tie back to individuals. And I could have, and we're calling it, in many cases, blended identities. I have my identity that I can validate biometrically, and I have this stable of identities, some of which may be ephemeral, some are going to be permanent agents.

The standing privilege thing has to go away. I think privilege access management's going to go away because everyone is now privileged. So the world has changed.

I used to do coding on 80-column cards. Hmm. I know.

I don't do that anymore because it doesn't- No. I'm not. You sure you don't?

Steven used it too. I could tell from the look on his face. That was before my time.

Yeah. That's true. I forget Steven's just a kid.

I think, I want to add a ray of optimism here, because I think, we have talked a lot about what can go wrong, how dangerous this situation is. I think there is a ray of optimism. And what I see is that this is a pendulum shift, right?

So everybody was excited about AI. Now the reality check, the production grade code and the return of investment is not matching. And it also gives some space for us to reflect on system thinking and software architecture.

This is AI debt. And how do you manage AI debt in an enterprise-grade organization? I work for critical infrastructure.

We cannot go back, right? We have to be very careful what we put into our systems as code. So now we have shifted our focus.

We have brought in system thinking view, the experimental nature of how and what AI could do is, of course, it's very interesting. But we have to also invest in this part of the business. And the second part, which is also what we were talking about, the larger ecosystem of open source and APIs and those kind of things.

I do believe that there will be some kind of AI common code or something like that, which would be the solution to this kind of a problem. Because we can't just go in zillion of directions with zillions of tools, right? So nobody will appreciate to invest that kind of time, energy, and money in hunting those bugs.

That is not sustainable in the longer run. So I feel that there is a ray of optimism that people are thinking about AI depth and how to cater to it. Agreed.

Agreed. Let's end on that positive note, though, Garima. I appreciate it.

We're about out of time. Garima, Jeff, it's great to have you back on. We haven't seen you enough on here.

I hope this will be the start of a weekly appearance or two. And Steven, it's great. I don't know when I'm going to see you in person, though, Steven.

We got to get back on the... Our paths keep missing. I know, right?

And one more thing that I just want to jump in here with. I am old enough to remember the "Spy vs Spy" cartoons in Mad Magazine. Mm-hmm.

Whenever anybody talks about agents versus agents, I always think of "Spy vs Spy," and I think about these two guys, you got the black hats and you got the white hats, and they're fighting each other. And basically, it's a stalemate because they have the same skills, they have the same tools, they're out there. And I always watched that and wondered, basically, if they both quit, they both wouldn't need to fight.

And I guess that's how it is with agentic AI, right? Too bad the rest of the world didn't work like that. Too bad we can't all get along.

Yeah. But we will be back tomorrow with more Techstrong gang. We've got Techstrong TV replay going on, I believe, after this.

Hey, all of our RSAC videos are up. Steven, are the Tech Field Day videos up yet, or shortly? Yeah, the Tech Field Day videos are up.

Commvault, Object First, and Veem, they should be in the app as well. They're on Techstrong TV, on the OTT app, and of course, on the Tech Field Day YouTube channel. Yeah.

So check them out there. Until tomorrow, guys. Have a great day.

This is Alan Schamel. We're out.