Heather Adkins

Yeah, in some ways it allows experimentation to be built on top of it. So meaning you can fine tune models because you have open weights and things of that sort.

Yeah, in some ways it allows experimentation to be built on top of it. So meaning you can fine tune models because you have open weights and things of that sort.

It's a little bit differently than the way that most have viewed open source in the past, which means think of it in a code context where you can see all of the lines of code and then compile it for yourself so that you can go ahead and investigate in terms of what influences were put into the actual code.

It's a little bit differently than the way that most have viewed open source in the past, which means think of it in a code context where you can see all of the lines of code and then compile it for yourself so that you can go ahead and investigate in terms of what influences were put into the actual code.

So in some ways I have to say the AI industry has been using the term open source rather loosely because it's fooling people into thinking that you can actually rebuild and recompile this thing for yourself, which you can't because you need the source training data. And that's where a lot of risks could be induced in these style of models. And therein lies the risk. What is it not showing us?

So in some ways I have to say the AI industry has been using the term open source rather loosely because it's fooling people into thinking that you can actually rebuild and recompile this thing for yourself, which you can't because you need the source training data. And that's where a lot of risks could be induced in these style of models. And therein lies the risk. What is it not showing us?

We don't know if it has embedded agents or not from a supply chain perspective. We just don't know. So there's certain things it's not revealing to us.

We don't know if it has embedded agents or not from a supply chain perspective. We just don't know. So there's certain things it's not revealing to us.

So without knowing what's an ingredient, I mean, you're eating the final product and basically at a cursory level, it's like taking a delivery of an eclair and saying, I can't really affect the creation of the eclair or recreate it, but I can change the icing on it. That's basically how you have to think about these things.

So without knowing what's an ingredient, I mean, you're eating the final product and basically at a cursory level, it's like taking a delivery of an eclair and saying, I can't really affect the creation of the eclair or recreate it, but I can change the icing on it. That's basically how you have to think about these things.

Right. It depends on how you're using it, whether as a local model or through their iOS app or through the API. So through the iOS app, I mean, that's literally, you know, it made it to the top of the app store.

Right. It depends on how you're using it, whether as a local model or through their iOS app or through the API. So through the iOS app, I mean, that's literally, you know, it made it to the top of the app store.

All these folks are downloading it and potentially compromising client or employee data as a result, not knowing that it's going to a hosted service and potentially even through contact centers that are monitoring what's flowing through the system.

All these folks are downloading it and potentially compromising client or employee data as a result, not knowing that it's going to a hosted service and potentially even through contact centers that are monitoring what's flowing through the system.

Governments are going and trying to hijack your accounts. They have very... front door kinds of ways that they can ask for that information. But here was a government clearly not going through the front door, clearly trying to find a workaround using hacking techniques. And that really did change everything for us.

Governments are going and trying to hijack your accounts. They have very... front door kinds of ways that they can ask for that information. But here was a government clearly not going through the front door, clearly trying to find a workaround using hacking techniques. And that really did change everything for us.

What we had thought were norms on the internet weren't actually norms.

What we had thought were norms on the internet weren't actually norms.

Google in 2009 was still a relatively small company, and the people who worked there had worked there for quite some time, right, and built Google, and also were of this generation of people who helped build the internet. And there's a certain philosophy of openness, connectedness, personal responsibility in how the internet was created, its culture.

Google in 2009 was still a relatively small company, and the people who worked there had worked there for quite some time, right, and built Google, and also were of this generation of people who helped build the internet. And there's a certain philosophy of openness, connectedness, personal responsibility in how the internet was created, its culture.

And that permeated through the culture at Google. And also people worked for Google in 2009 because they really believed the mission. organize all the world's information, make it universally accessible. They could see the information revolution online, how it connected people. And the idea that someone would want to violate that, I think, really spoke very strongly to Googlers.

And that permeated through the culture at Google. And also people worked for Google in 2009 because they really believed the mission. organize all the world's information, make it universally accessible. They could see the information revolution online, how it connected people. And the idea that someone would want to violate that, I think, really spoke very strongly to Googlers.

Like, this is a boundary. Why did you cross this boundary? Googlers really saw this as a shocking moment. Like, I can't believe somebody went there and did that. I can't believe a government went there and did that. And people really took that to heart.

Like, this is a boundary. Why did you cross this boundary? Googlers really saw this as a shocking moment. Like, I can't believe somebody went there and did that. I can't believe a government went there and did that. And people really took that to heart.

I will say that it switched on a kind of strange paranoia. I remember, and this could have just been the sleep deprivation, but I remember driving into work one morning, it was very early, and I saw a telco truck, a telecoms truck, in the middle of the road.

I will say that it switched on a kind of strange paranoia. I remember, and this could have just been the sleep deprivation, but I remember driving into work one morning, it was very early, and I saw a telco truck, a telecoms truck, in the middle of the road.

They'd coned everything off, and they'd had the manhole cover open, and it was right next to campus, and I thought, what if they're tapping the fiber, right? There's this weird paranoia that kicks in. That I think that kind of thinking did come, you know, not rational, but I think you do start to question everything you see and all the decisions you're making.

They'd coned everything off, and they'd had the manhole cover open, and it was right next to campus, and I thought, what if they're tapping the fiber, right? There's this weird paranoia that kicks in. That I think that kind of thinking did come, you know, not rational, but I think you do start to question everything you see and all the decisions you're making.

So we set up these honeypots in the hopes that we would learn which other corporate laptops we're going to connect in. And via there, we were able to see the other victims and triangulate who they were.

So we set up these honeypots in the hopes that we would learn which other corporate laptops we're going to connect in. And via there, we were able to see the other victims and triangulate who they were.

I would say there were one or two companies where we called them up and said, you know, hey, you've got this thing. And they said, yeah, we've been having trouble with that for a while.

I would say there were one or two companies where we called them up and said, you know, hey, you've got this thing. And they said, yeah, we've been having trouble with that for a while.

Thank you.

Thank you.

It was kind of the end of the day, the end of my working day, and I'd come back to my desk and there were a bunch of engineers standing around a desk and said, Haley, look what we found. We did actually think... What we were seeing, you know, was an intern just with bad business practice.

It was kind of the end of the day, the end of my working day, and I'd come back to my desk and there were a bunch of engineers standing around a desk and said, Haley, look what we found. We did actually think... What we were seeing, you know, was an intern just with bad business practice.

You know, once we'd spent a few hours taking a look at it, it was pretty obvious that what we were looking at was very different.

You know, once we'd spent a few hours taking a look at it, it was pretty obvious that what we were looking at was very different.

I would say we did every investigative method you could think of, from forensics to interviews. If a person's machine or an account was used by the threat actor, we interviewed them. We did everything. So yes, we would have talked to interns, Googlers of all kinds.

I would say we did every investigative method you could think of, from forensics to interviews. If a person's machine or an account was used by the threat actor, we interviewed them. We did everything. So yes, we would have talked to interns, Googlers of all kinds.

all of the systems administrators whose accounts you know the sre's we call them whose accounts were taken over and abused by the threat actor we talked to them i think i even made two people change their username on the systems just so that we could delineate between good guy bad guy kind of thing i would say we deployed all creative resources after the first 12 to 24 hours it was

all of the systems administrators whose accounts you know the sre's we call them whose accounts were taken over and abused by the threat actor we talked to them i think i even made two people change their username on the systems just so that we could delineate between good guy bad guy kind of thing i would say we deployed all creative resources after the first 12 to 24 hours it was

pretty clear that we were dealing with a scale that was going to quickly overwhelm our small team who knew how to do this work. So very quickly we knew we needed to call in people who were doing this on a regular basis.

pretty clear that we were dealing with a scale that was going to quickly overwhelm our small team who knew how to do this work. So very quickly we knew we needed to call in people who were doing this on a regular basis.

When they made the call, the Googlers only had one rule. Don't wear suits. Don't show up in suits, right? If you at the time had shown up at Google in a suit, People would have thought like the FBI was here and hanging out with the security team, that's probably not a good sign. And sure enough, they did come in suits and had to buy clothes for the rest of the engagement on site.

When they made the call, the Googlers only had one rule. Don't wear suits. Don't show up in suits, right? If you at the time had shown up at Google in a suit, People would have thought like the FBI was here and hanging out with the security team, that's probably not a good sign. And sure enough, they did come in suits and had to buy clothes for the rest of the engagement on site.

But I think, you know, I have a really clear memory of that day when they came because we gathered, you know, a handful of us in a conference room and we briefed them on what we had seen. And, you know, it was immediate. They were like, yes. We know exactly what this is. I don't know that we've ever seen this particular threat actor before, but this looks just like Chinese APT.

But I think, you know, I have a really clear memory of that day when they came because we gathered, you know, a handful of us in a conference room and we briefed them on what we had seen. And, you know, it was immediate. They were like, yes. We know exactly what this is. I don't know that we've ever seen this particular threat actor before, but this looks just like Chinese APT.

I think that that moment, it was nice to have experienced adults in the room. It felt like we had a really solid partner. So I've forgiven them for wearing suits.

I think that that moment, it was nice to have experienced adults in the room. It felt like we had a really solid partner. So I've forgiven them for wearing suits.

As we dug in, it became clear actually that we think it was, the whole attack was about long-term access to Gmail accounts.

As we dug in, it became clear actually that we think it was, the whole attack was about long-term access to Gmail accounts.