Latest / a16z AI Policy Brief / Securing America’s Infrastructure With AI

Transcript

Securing America’s Infrastructure With AI · a16z AI Policy Brief

0:00
SPEAKER_00: I think the reality is that the economics of cybersecurity was just broken before AI.
0:05
SPEAKER_00: AI allows attackers to essentially jiggle every digital doorknob continuously.
0:10
SPEAKER_00: And at school,
0:11
SPEAKER_00: AI can take that data,
0:13
SPEAKER_00: crunch it far more quickly,
0:15
SPEAKER_00: and identify what's a real threat and potentially block it.
0:20
SPEAKER_01: Every application of something good produces a risk,
0:23
SPEAKER_01: and we just need to manage the opportunity and the risk as kind of one and the same
0:29
SPEAKER_01: in a way.
0:30
SPEAKER_00: It brings real benefits.
0:32
SPEAKER_00: City leaders can sleep well at night knowing a school system won't get hacked,
0:36
SPEAKER_00: children's records stolen,
0:37
SPEAKER_00: knowing a local hospital won't be disrupted, that peace of mind.
0:40
SPEAKER_01: There are ways that we can help them understand and apply the lessons to improve
0:46
SPEAKER_01: the benefits and reduce the risks.
0:48
SPEAKER_02: A small town shouldn't need the technology budget of a Fortune 500 company
0:53
SPEAKER_02: to protect its essential services from cyber attacks.
0:56
SPEAKER_02: I'm Jay Ramaswamy,
0:57
SPEAKER_02: and today I'm joined by Ann Neuberger and Steven Goldsmith,
1:00
SPEAKER_02: who will introduce Securing America's Infrastructure Foundation, or SAFE.
1:05
SPEAKER_02: SAFE is a new nonprofit with the goal of helping communities
1:08
SPEAKER_02: with limited resources use AI to protect their essential services from cyber attack.
1:13
SPEAKER_02: Many small cities, schools, and hospitals can't afford the tools and expertise they need.
1:19
SPEAKER_02: SAFE will provide technology, hands-on support, and training at no cost.
1:25
SPEAKER_02: A16z general partner and head of global affairs Ann Neuberger will be leading
1:30
SPEAKER_02: the foundation initially as a volunteer.
1:32
SPEAKER_02: Stephen Goldsmith is one of SAFE's founding board members.
1:36
SPEAKER_02: Stephen is a former mayor of Indianapolis and deputy mayor of New York City.
1:40
SPEAKER_02: He now teaches urban policy at Harvard's Kennedy School and leads the Data Smart City Solutions.
1:47
SPEAKER_02: I hope you enjoy the conversation.
1:51
SPEAKER_01: And Stephen, thanks for joining me today.
1:53
SPEAKER_02: And can we just start with the big picture and have you tell us about the goals of
1:57
SPEAKER_02: the foundation and why it's being launched at this moment in time?
2:00
SPEAKER_02: And finally, what you hope to achieve.
2:03
SPEAKER_00: Absolutely.
2:04
SPEAKER_00: So in a really practical way,
2:07
SPEAKER_00: We know that AI brings a lot of innovation and can really make a difference in
2:13
SPEAKER_00: the way municipalities serve citizens,
2:16
SPEAKER_00: in the way municipalities can secure really critical systems.
2:19
SPEAKER_00: We also know AI brings risks.
2:21
SPEAKER_00: When I served in the White House,
2:22
SPEAKER_00: there was a wave of ransomware-driven cyberattacks,
2:25
SPEAKER_00: and we can expect that those same criminals will use AI to further drive attacks.
2:29
SPEAKER_00: So on a very practical level, we want to help parts of America use AI to
2:38
SPEAKER_00: That's super interesting.
2:47
SPEAKER_02: Steven, you're currently director of Data Smart City Solutions at the JFK School of Government.
2:54
SPEAKER_02: You were previously mayor of Indianapolis and the deputy mayor of New York City.
2:59
SPEAKER_02: What is it really like in those seats trying to bring innovation
3:03
SPEAKER_02: to municipal services and critical infrastructure?
3:06
SPEAKER_02: What have you learned about engaging local governments
3:09
SPEAKER_02: and critical infrastructure operators to address the concerns
3:12
SPEAKER_02: that municipal governments have?
3:15
SPEAKER_01: Let's start with innovation generally.
3:17
SPEAKER_01: I mean, governments, particularly, I worked for Mike Bloomberg in New York City.
3:21
SPEAKER_01: That was a well-run government.
3:23
SPEAKER_01: But well-run meant it reliably did the same thing day in and day out.
3:30
SPEAKER_01: Big, bold innovation, which injects disruptive change,
3:37
SPEAKER_01: So one thing we know about innovation is it's disruptive
3:40
SPEAKER_01: and it creates changes throughout the organization.
3:43
SPEAKER_01: In the issues we're talking about today with critical infrastructure,
3:47
SPEAKER_01: software, and AI,
3:49
SPEAKER_01: that one of the challenges is there is a department that owns that responsibility.
3:55
SPEAKER_01: But that's an outdated model, right?
3:56
SPEAKER_01: Because the risk and benefit is distributed throughout the organization.
4:00
SPEAKER_01: So I think the best way to summarize and answer to your questions is
4:04
SPEAKER_01: to say innovation is disruptive,
4:07
SPEAKER_01: technology innovation is particularly disruptive,
4:10
SPEAKER_01: both because of its impact and its implications,
4:13
SPEAKER_01: and we need to change the model of government where a department,
4:17
SPEAKER_01: an organization,
4:19
SPEAKER_01: or in a small town,
4:20
SPEAKER_01: one person kind of owns the process of implementing and the process of reducing
4:27
SPEAKER_01: the risk.
4:29
SPEAKER_02: What do you think are the challenges facing that sort of reorganization?
4:33
SPEAKER_02: And how can the foundation assist in helping governments get to where they need
4:39
SPEAKER_02: to get to, to have the right structure to address these complex issues?
4:45
SPEAKER_01: If you go back to Anne's answer to your question,
4:49
SPEAKER_01: so what AI does, potentially, is unlocks enormous power in city government.
4:58
SPEAKER_01: And it does it by allowing individuals who previously had no access to using data
5:05
SPEAKER_01: to be able to use data to be able to enhance their discretion.
5:10
SPEAKER_01: So broadly increasing government productivity means that we're broadly increasing risks, right?
5:20
SPEAKER_01: IoT devices, every individual doing different things, even with inside regulation.
5:28
SPEAKER_01: Relate to the ability to have more employees in a public organization be responsive,
5:35
SPEAKER_01: understand the needs of their community,
5:37
SPEAKER_01: use data to do predictive and preemptive work, right,
5:40
SPEAKER_01: to identify problems before they occur,
5:42
SPEAKER_01: to be responsive to human needs in the way they never were before.
5:47
SPEAKER_01: But every one of those touches produces a risk.
5:54
SPEAKER_02: And Stephen's given us the ground level view, sort of the frontline view of what's going on.
6:00
SPEAKER_02: You had a different view,
6:01
SPEAKER_02: also frontline,
6:02
SPEAKER_02: but a little bit more of a 50,000 foot view,
6:05
SPEAKER_02: you know,
6:06
SPEAKER_02: from leading intelligence operations,
6:08
SPEAKER_02: establishing and leading the NSA's cyber organization.
6:13
SPEAKER_02: Could you share with the audience,
6:15
SPEAKER_02: what did those experiences teach you about how the foundation will focus
6:20
SPEAKER_02: on securing critical infrastructure,
6:22
SPEAKER_02: both the concerns that Stephen has raised,
6:25
SPEAKER_02: but also other ones that you experienced when you were in government?
6:29
SPEAKER_00: Yeah, you know, it's so interesting.
6:30
SPEAKER_00: When I moved from NSA to the White House,
6:34
SPEAKER_00: had somebody asked me where I'd spend a large amount of time,
6:39
SPEAKER_00: I never would have expected that it would have been Responding to disruption
6:43
SPEAKER_00: of infrastructure due to cyber attacks.
6:46
SPEAKER_00: You know, Colonial Pipeline was the first one.
6:48
SPEAKER_00: Three months after President Biden took office,
6:51
SPEAKER_00: it disrupted the only pipeline that serves the entire East Coast of America.
6:55
SPEAKER_00: And it touched Americans' lives instantly.
6:57
SPEAKER_00: Right.
6:57
SPEAKER_00: You had cars queued up at gas stations.
7:00
SPEAKER_00: Right.
7:01
SPEAKER_00: Because people couldn't have gas.
7:03
SPEAKER_00: Similarly,
7:03
SPEAKER_00: you know,
7:04
SPEAKER_00: one of the cyber attacks that I just remember most vividly was a cyber attack
7:09
SPEAKER_00: against a hospital chain called Ascension Healthcare.
7:13
SPEAKER_00: And it disrupted the operations of the hospital.
7:15
SPEAKER_00: The CEO said to me, Anne,
7:16
SPEAKER_00: I had to hire runners to run images from MRI into surgery because
7:21
SPEAKER_00: the network was locked.
7:22
SPEAKER_00: The network was encrypted.
7:24
SPEAKER_00: And I recall sitting in the White House,
7:26
SPEAKER_00: you know,
7:26
SPEAKER_00: when I talked to the secretary of HHS and the deputy,
7:28
SPEAKER_00: and I said, OK, how can we be most helpful?
7:31
SPEAKER_00: The deputy secretary,
7:32
SPEAKER_00: Andrea Palm,
7:32
SPEAKER_00: said to me, Anne,
7:34
SPEAKER_00: the first thing we need to figure out is which are the communities where
7:38
SPEAKER_00: a hospital was hit, and it's the community's only hospital.
7:42
SPEAKER_00: Meaning if there's two hospitals 10 minutes apart from each other, ambulances are rerouted.
7:47
SPEAKER_00: As some of us may have seen in the pit,
7:48
SPEAKER_00: it causes real impact in that hospital,
7:50
SPEAKER_00: but the community still has service for emergency services.
7:54
SPEAKER_00: If it's the community's only hospital, then potentially that has an impact on lives.
7:59
SPEAKER_00: And that's one of the reasons why one of the efforts I'd like to work on early
8:05
SPEAKER_00: on is America's 1,800 rural hospitals.
8:08
SPEAKER_00: Those are the only hospitals in those rural communities.
8:12
SPEAKER_00: Many hospitals have closed in the last number of years due to costs, due to various things.
8:17
SPEAKER_00: Those hospitals lack the skills,
8:19
SPEAKER_00: the people, the money to ensure that they are not disrupted due to cyber attacks.
8:26
SPEAKER_00: As Stephen mentioned,
8:27
SPEAKER_00: AI makes a real difference in being able to economically defend networks,
8:31
SPEAKER_00: to find vulnerabilities,
8:33
SPEAKER_00: to review the code of MRI machines, generate patches for those.
8:36
SPEAKER_00: So we want to make an early target.
8:39
SPEAKER_00: A16z.
9:01
SPEAKER_00: Adopt the model that actually New York City and New York State used,
9:05
SPEAKER_00: where they set up a security operations center that was able
9:08
SPEAKER_00: to support multiple smaller communities in responding to an incident.
9:12
SPEAKER_00: So those are the three parts we're thinking about
9:15
SPEAKER_00: in helping critical infrastructure leverage AI to secure
9:18
SPEAKER_00: the critical systems Americans rely on every day.
9:21
SPEAKER_02: Great.
9:22
SPEAKER_02: Let's go a little bit deeper into the impact of AI on small cities,
9:27
SPEAKER_02: on schools, and on hospitals, as you mentioned.
9:31
SPEAKER_02: Can you be a little bit more granular in how AI impacts the threats
9:37
SPEAKER_02: to critical infrastructure that Americans rely on every day?
9:40
SPEAKER_02: And I'm thinking about, you know, what can people do to defend themselves against these threats?
9:46
SPEAKER_02: But also,
9:47
SPEAKER_02: what are the kinds of things that a school needs to be worried about or a city needs
9:51
SPEAKER_02: to be worried about?
9:52
SPEAKER_02: And Anne, why don't we start with you?
9:55
SPEAKER_00: Absolutely.
9:56
SPEAKER_00: You know,
9:57
SPEAKER_00: we used to say that attackers,
9:59
SPEAKER_00: it's always easier to be an attacker than a defender because an attacker has
10:02
SPEAKER_00: to find one digital door or window ajar.
10:05
SPEAKER_00: And a defender has to be monitoring all of them and really having visibility on all
10:12
SPEAKER_00: of them and ensuring that they're properly defending.
10:15
SPEAKER_00: AI allows attackers to essentially jiggle every digital doorknob continuously.
10:21
SPEAKER_00: So it just, it really does make it easier for attackers to conduct attacks.
10:26
SPEAKER_00: And the ones I'm thinking of mostly are criminals who do this for financial gain,
10:30
SPEAKER_00: let alone countries that are often more sophisticated and AI just ups their skill.
10:36
SPEAKER_00: But it does,
10:37
SPEAKER_00: AI does enable criminals who just want to make a buck by locking up a network
10:41
SPEAKER_00: and asking for a ransom to be paid.
10:44
SPEAKER_00: On the flip side,
10:47
SPEAKER_00: the hardest part of what makes cybersecurity sometimes
10:50
SPEAKER_00: so challenging is large amounts of data and figuring out signal from noise,
10:55
SPEAKER_00: not generating so many false alerts that,
10:58
SPEAKER_00: you know,
10:59
SPEAKER_00: a team of people or the often one individual responsible
11:02
SPEAKER_00: for cybersecurity spends their time going through what's a real alert versus
11:07
SPEAKER_00: a fake alert.
11:08
SPEAKER_00: AI can take that data,
11:10
SPEAKER_00: crunch it far more quickly,
11:11
SPEAKER_00: and identify what's a real threat and potentially block it.
11:17
SPEAKER_00: You know, not waiting for the person.
11:19
SPEAKER_00: If it has a question, block it and alert for a person to then look at that.
11:24
SPEAKER_00: That's one key area.
11:25
SPEAKER_00: The second key area is a lot of code.
11:28
SPEAKER_00: The industry uses a term called tech debt.
11:34
SPEAKER_00: A lot of old systems that weren't updated,
11:37
SPEAKER_00: that weren't upgraded,
11:38
SPEAKER_00: that were built before cyber attacks were as much of a part of life
11:42
SPEAKER_00: as they are today.
11:44
SPEAKER_00: AI makes it far easier to look at that code,
11:49
SPEAKER_00: find vulnerabilities,
11:50
SPEAKER_00: and generate a patch that a human still has to review, but then to secure that code.
11:55
SPEAKER_00: And that makes a big difference in critical infrastructure, power,
11:58
SPEAKER_00: water systems,
11:59
SPEAKER_00: hospitals that often didn't have the money to upgrade their systems over the years
12:03
SPEAKER_00: and have a lot of old systems that are,
12:06
SPEAKER_00: you know, blood in the water for cyber sharks.
12:15
SPEAKER_02: If it existed at the time and at the scale and competence that it does today,
12:19
SPEAKER_02: how would that have affected your ability sitting as Deputy Mayor to address some
12:24
SPEAKER_02: of these challenges?
12:26
SPEAKER_01: Well, we're going to assume that every one of your questions means what are
12:31
SPEAKER_01: the potential benefits and what are the potential liabilities?
12:36
SPEAKER_01: Because, you know, it's silly to...
12:39
SPEAKER_01: Assume that this question is a one-sided coin, right?
12:42
SPEAKER_01: So every application of something good produces a risk.
12:45
SPEAKER_01: And we just need to manage the opportunity and the risk as kind of one and the same in a way.
12:52
SPEAKER_01: So I've spent a fair amount of time on technology innovation in cities in a range
12:59
SPEAKER_01: of ways, including in New York City where part of my portfolio was IT.
13:06
SPEAKER_01: I think what we see today, though,
13:09
SPEAKER_01: is the proliferation of sensors and the proliferation of handheld devices, right?
13:16
SPEAKER_01: So every citizen, every resident in a city is a sensor and a user, right?
13:24
SPEAKER_01: And there are IoT devices everywhere,
13:28
SPEAKER_01: traffic signals to how you manage the gates and water systems and the like.
13:33
SPEAKER_01: So the ability to manage discreetly,
13:36
SPEAKER_01: precisely, the ability for precision government is dramatically better.
13:41
SPEAKER_01: But there's very little consideration of how those sources of information into
13:47
SPEAKER_01: the city can,
13:49
SPEAKER_01: particularly with AI looking for vulnerabilities, Can produce distortions.
13:55
SPEAKER_01: So what we see is broadly customized use, precision use of the data.
14:01
SPEAKER_01: Think just traffic signals.
14:03
SPEAKER_01: How many times have you complained about,
14:05
SPEAKER_01: I'm sitting this red light and there's nobody coming down through the green light.
14:10
SPEAKER_01: Why don't they just change this light?
14:12
SPEAKER_01: Well, it's good.
14:12
SPEAKER_01: They should.
14:13
SPEAKER_01: They should change every light in real time in every intersection in every city tomorrow.
14:18
SPEAKER_01: That would make Greatly enhanced services,
14:21
SPEAKER_01: a benefit through AI adjusting the signals, and then every signal is at risk, right?
14:27
SPEAKER_01: So you just have to figure out how to manage the risk as you manage the benefit.
14:31
SPEAKER_02: And in some senses, Hollywood beat us to the punch here.
14:34
SPEAKER_02: I remember a Fast and Furious movie with the god's eye,
14:37
SPEAKER_02: which describes exactly what you're describing.
14:40
SPEAKER_02: Following up a little bit on some of the issues that you've raised,
14:45
SPEAKER_02: you wrote a piece for Foreign Affairs that really gets at the core of why AI is such
14:49
SPEAKER_02: an important defense tool.
14:53
SPEAKER_02: Namely, it's bringing down the cost of defending infrastructure.
14:59
SPEAKER_02: What does AI actually change about defending a small city or a hospital?
15:05
SPEAKER_02: And how does that cost equation kind of fit into this whole picture?
15:08
SPEAKER_00: I think the reality is that the economics of cybersecurity was just broken before AI.
15:14
SPEAKER_00: We'd see it because we saw this wave of ransomware attacks really driven by The rise
15:22
SPEAKER_00: of ways that people could actually pay a ransom.
15:24
SPEAKER_00: They could pay via crypto.
15:25
SPEAKER_00: So as a result, that kind of enabled this borderless targeting.
15:29
SPEAKER_00: You could have attackers in one country using infrastructure in a second set
15:34
SPEAKER_00: of countries, targeting hospitals and pipelines in a third.
15:39
SPEAKER_00: And the economics,
15:40
SPEAKER_00: because many of these systems were old,
15:43
SPEAKER_00: because many of smaller critical infrastructure across the country,
15:46
SPEAKER_00: particularly more rural parts of the country,
15:49
SPEAKER_00: Couldn't necessarily attract people or pay people
15:52
SPEAKER_00: as many organizations wanted top cyber talent.
15:57
SPEAKER_00: And as a result,
15:59
SPEAKER_00: what happened is you had a lot of organizations who felt forced to either pay
16:04
SPEAKER_00: a ransom and couldn't, you know, kind of were living with that level of risk.
16:20
SPEAKER_00: Anomalous activity, find malicious activity, and block it.
16:24
SPEAKER_00: There's also real open source capabilities, which make it far less costly.
16:30
SPEAKER_00: It's cheap to actually deploy.
16:32
SPEAKER_00: As well as the code capabilities.
16:34
SPEAKER_00: The code capabilities we all know about in terms
16:36
SPEAKER_00: of generating new applications also enable finding vulnerabilities
16:41
SPEAKER_00: and generating new code to patch those in a significant way.
16:45
SPEAKER_00: And that helps with a lot of these old systems
16:47
SPEAKER_00: that Organizations can't necessarily afford to replace,
16:52
SPEAKER_00: but are often internet connected.
16:53
SPEAKER_00: Think about an MRI machine.
16:54
SPEAKER_00: It needs remote access to the internet because it's not going to have
16:57
SPEAKER_00: a full-time maintenance engineer there.
17:00
SPEAKER_00: So they'll often have remote access to dial in and fix it if there's an issue.
17:03
SPEAKER_00: That dial in is often, or that remote connection, is often not secure.
17:09
SPEAKER_00: As a result,
17:10
SPEAKER_00: you know,
17:10
SPEAKER_00: when you can secure that, first,
17:13
SPEAKER_00: when you can find the vulnerabilities in the MRI code and fix it,
17:16
SPEAKER_00: But then also secure the login, you fundamentally close two doors.
17:22
SPEAKER_02: I noticed you used the word dial-in,
17:24
SPEAKER_02: and I understood exactly what you meant,
17:26
SPEAKER_02: but I'm sure a number of our audience were probably like, what exactly was dial-in?
17:30
SPEAKER_02: But I remember those days all too well as well.
17:32
SPEAKER_00: That's how many systems are, right?
17:35
SPEAKER_00: Or in some cases, update the remote system.
17:39
SPEAKER_02: Absolutely.
17:41
SPEAKER_02: Could you talk a little bit about digital twins and how that might be
17:45
SPEAKER_02: an important component of what these new technologies can bring to us
17:50
SPEAKER_02: and how they might help a city prioritize,
17:52
SPEAKER_02: for example, what exactly they should be addressing first in the event of an attack?
18:10
SPEAKER_00: Many of these systems today are 24-7 systems.
18:13
SPEAKER_00: Americans rely on them every single day.
18:15
SPEAKER_00: And yet,
18:17
SPEAKER_00: Actually,
18:18
SPEAKER_00: using AI to find and fix vulnerabilities,
18:22
SPEAKER_00: to find and fix misconfigurations would be ideal.
18:25
SPEAKER_00: But engineers,
18:27
SPEAKER_00: IT engineers,
18:28
SPEAKER_00: operators,
18:29
SPEAKER_00: CISOs are worried about just either applying patches,
18:33
SPEAKER_00: which vendors like Cisco or Microsoft have been generating.
18:36
SPEAKER_00: I think we've all seen the hockey stick number of patches
18:39
SPEAKER_00: that have been generated just by tech developers.
18:43
SPEAKER_00: So many of the companies operating power systems pipelines are really worried
18:49
SPEAKER_00: about applying that massive new volume of patches or,
18:52
SPEAKER_00: frankly, using AI to actively hunt and then generate fixes for their network.
18:59
SPEAKER_00: Because these are live networks that can't afford any downtime.
19:02
SPEAKER_00: The concept of digital twins is essentially a virtual replica,
19:06
SPEAKER_00: which AI can now generate today at different levels of specificity.
19:11
SPEAKER_00: At a high level of specificity,
19:12
SPEAKER_00: a digital replica,
19:14
SPEAKER_00: enough that you can test a patch and see are there second order effects,
19:18
SPEAKER_00: test a configuration change and see what other thing it may trigger.
19:23
SPEAKER_00: And as with many AI systems, these systems can learn over time.
19:27
SPEAKER_00: So by testing fixes, A,
19:31
SPEAKER_00: you help build a virtual replica that's more and more precise to that network,
19:34
SPEAKER_00: and B, you can then see Will there be second-order effects?
19:38
SPEAKER_00: And essentially test it before you then roll out those AI-driven cybersecurity fixes
19:43
SPEAKER_00: on the live network.
19:44
SPEAKER_00: So in that piece I wrote,
19:46
SPEAKER_00: I was taking to your earlier point that big-picture view of America as
19:49
SPEAKER_00: a country has massive amounts of critical infrastructure risk
19:54
SPEAKER_00: because it's far easier to attack than defend.
19:56
SPEAKER_00: And we have adversaries,
19:59
SPEAKER_00: competitors,
20:00
SPEAKER_00: the governments of Russia, China, Iran, who have very capable cyber programs.
20:05
SPEAKER_00: And the U.S.
20:06
SPEAKER_00: government needs a way to figure out what are the most critical threats that as
20:10
SPEAKER_00: a country we must find in lockdown.
20:12
SPEAKER_00: Because these are live systems,
20:14
SPEAKER_00: many are worried about just using AI to actively and directly apply those fixes.
20:20
SPEAKER_00: So building digital replicas can help us have a kind of a bridge path to do that.
20:28
SPEAKER_02: And Stephen...
20:30
SPEAKER_02: What do you see as the most critical or maybe most,
20:35
SPEAKER_02: if there are more than one,
20:36
SPEAKER_02: critical bottlenecks to getting the benefits from this technology?
20:42
SPEAKER_02: Is it budget?
20:43
SPEAKER_02: We all know that cities and towns and hospitals are strapped.
20:47
SPEAKER_02: Is it technical expertise?
20:49
SPEAKER_02: Is it something else?
20:51
SPEAKER_02: But from your experiences,
20:54
SPEAKER_02: what could be potential bottlenecks and preventative Things that would prevent us
21:01
SPEAKER_02: from realizing the full potential of this technology.
21:05
SPEAKER_01: Well, certainly everything you just said is a bottleneck.
21:09
SPEAKER_01: I don't think I could pick just one.
21:11
SPEAKER_01: I mean, so if we step back,
21:13
SPEAKER_01: particularly looking at smaller jurisdictions,
21:16
SPEAKER_01: one bottleneck is they all need Anne's advice,
21:19
SPEAKER_01: and she's probably unavailable to help every community in the U.S., right?
21:22
SPEAKER_01: So we have an expertise problem that's pretty significant.
21:27
SPEAKER_01: And I don't mean just because the cities are small.
21:29
SPEAKER_01: I mean in every city in America,
21:31
SPEAKER_01: there is an expertise deficit because the technology is moving so quickly.
21:36
SPEAKER_01: And often city IT teams are not quite as
21:42
SPEAKER_01: avant-garde as the technology is.
21:45
SPEAKER_01: So there's a lapse in time.
21:48
SPEAKER_01: So you've got a technology issue.
21:51
SPEAKER_01: And to a large extent,
21:53
SPEAKER_01: I think we need to rethink The kind of shared service model of what can be provided
22:00
SPEAKER_01: in terms of expertise to a community that's not on site.
22:03
SPEAKER_01: Because it's not practical to assume that the small hospital or small city or even
22:09
SPEAKER_01: a large city is going to have the requisite knowledge.
22:12
SPEAKER_01: So that's one set of things.
22:14
SPEAKER_01: The second is that there is difficulty purchasing services.
22:21
SPEAKER_01: And critical infrastructure,
22:22
SPEAKER_01: because that means you have to know enough to be able to buy the right amount
22:29
SPEAKER_01: that you can afford.
22:30
SPEAKER_01: So you said, is there a budget issue?
22:32
SPEAKER_01: Of course, there's a budget issue.
22:33
SPEAKER_01: But inside that budget issue is, okay, I have a budget.
22:36
SPEAKER_01: How do I spend it for the best return on investment?
22:38
SPEAKER_01: That assumes that there's the expertise that I just talked about necessary to actually...
22:44
SPEAKER_01: So one of the issues we see in cities today,
22:45
SPEAKER_01: and I've been talking to many of the top 40, is that...
22:50
SPEAKER_01: How do they purchase,
22:52
SPEAKER_01: not even critical infrastructure, how do they purchase cloud software that uses AI?
22:59
SPEAKER_01: How will they evaluate the risks of the AI in the vendor's solution?
23:03
SPEAKER_01: And they don't know how to do that.
23:05
SPEAKER_01: I don't mean it as a criticism.
23:06
SPEAKER_01: It's just very difficult to do.
23:08
SPEAKER_01: So we have a budget issue.
23:09
SPEAKER_01: We have an expertise issue.
23:10
SPEAKER_01: I think we have a shared service model issue, which is the model for protection is outdated.
23:16
SPEAKER_01: All of those things.
23:17
SPEAKER_01: And we then finally,
23:18
SPEAKER_01: I guess,
23:18
SPEAKER_01: have maybe perhaps more easily addressed as just an education issue,
23:23
SPEAKER_01: which is how does an employee who is a caseworker or utility worker
23:30
SPEAKER_01: a hospital worker using some IT that has AI in it,
23:35
SPEAKER_01: how do we train them to be more cautious?
23:38
SPEAKER_00: Yeah.
23:40
SPEAKER_00: I would just add,
23:40
SPEAKER_00: if I could, Jay,
23:41
SPEAKER_00: I think Stephen's point,
23:43
SPEAKER_00: it was implicit in everything he said, there's sometimes also a trust issue.
23:48
SPEAKER_00: You know, such efforts haven't always succeeded in the past.
23:51
SPEAKER_00: It's something we talked about quite a bit before in terms of,
23:56
SPEAKER_00: because sometimes there were efforts that said,
23:58
SPEAKER_00: we'll come in, we'll deploy some new tech, best wishes.
24:01
SPEAKER_00: And the recipient may not necessarily have The time,
24:06
SPEAKER_00: the skill, or the resources to actually keep that updated.
24:10
SPEAKER_00: And that's why one of the things we've been thinking about,
24:12
SPEAKER_00: and Stephen just referenced it,
24:13
SPEAKER_00: so first is building trust is critical,
24:16
SPEAKER_00: that we're engaged, you know, the model has a long-term model.
24:21
SPEAKER_00: So, you know,
24:21
SPEAKER_00: one of the models that I talked about quite a bit
24:25
SPEAKER_00: with Colin Ahern is now Governor Hochul of New York's Cyber Advisor,
24:30
SPEAKER_00: but he previously was the head of New York City Cyber Command.
24:34
SPEAKER_00: And one of the things they built was when New York City stood up,
24:37
SPEAKER_00: which clearly is a city with both a target of threats,
24:40
SPEAKER_00: but also real resources, the ability to attract really skilled people.
24:45
SPEAKER_00: Stephen knows this better than anyone.
24:48
SPEAKER_00: And what they recognized after they stood up a security operations center is
24:52
SPEAKER_00: by economies of scale, they had extra scale.
24:55
SPEAKER_00: They had extra resources there.
24:57
SPEAKER_00: And then like a smaller municipality called Suffolk got hit by a cyber attack and
25:02
SPEAKER_00: And, you know, all city services were disrupted, networks were completely locked up.
25:07
SPEAKER_00: And what New York City realized was that if they built a standard package
25:11
SPEAKER_00: for smaller municipalities and said, guys,
25:13
SPEAKER_00: you deploy this standards monitoring package,
25:17
SPEAKER_00: the data will come to our central security operations center where we have the scale
25:21
SPEAKER_00: and we'll be constantly evaluating the data.
25:25
SPEAKER_00: And that was a cool model because it leveraged the return on investment for
25:29
SPEAKER_00: the resource investment and helped the bigger city help smaller cities in a way
25:34
SPEAKER_00: where,
25:35
SPEAKER_00: you know,
25:36
SPEAKER_00: the small marginal cost,
25:37
SPEAKER_00: because the fixed cost of standing up a SOC,
25:38
SPEAKER_00: a security operations center,
25:40
SPEAKER_00: was there whether it was, you know, one network or five.
25:54
SPEAKER_00: You know, you have a small team who can potentially support that.
25:57
SPEAKER_00: We'll need to figure out the details of what that looks like,
25:59
SPEAKER_00: but that's, you know, giving credit to New York for innovating on that.
26:03
SPEAKER_00: I think that's the model we have in mind for these kinds of places.
26:08
SPEAKER_02: In addition to that sort of novel and innovative model,
26:12
SPEAKER_02: are there other things that SAFE will be doing differently
26:16
SPEAKER_02: that can impact what success looks like in this kind of endeavor?
26:22
SPEAKER_00: I think first,
26:23
SPEAKER_00: you know,
26:24
SPEAKER_00: our focus is on learning from prior efforts to see,
26:28
SPEAKER_00: you know, where did they go right?
26:30
SPEAKER_00: Where did they go wrong?
26:31
SPEAKER_00: We really feel AI,
26:34
SPEAKER_00: third part,
26:34
SPEAKER_00: is the game changer because AI does make cybersecurity far easier
26:39
SPEAKER_00: while marrying it up with models,
26:41
SPEAKER_00: like I mentioned,
26:42
SPEAKER_00: where what we do is we try to deploy what is standard
26:47
SPEAKER_00: in common places like hospitals,
26:49
SPEAKER_00: add the resources of Centralizing some things so that the degree of local training
26:57
SPEAKER_00: and local ongoing responsibility could potentially be minimized and managed,
27:01
SPEAKER_00: recognizing that for a small municipality or a small pipeline or water system,
27:08
SPEAKER_00: They're not going to be able to afford a full-time cybersecurity person,
27:13
SPEAKER_00: or there are real benefits to the city from using AI, for example, to find potholes.
27:21
SPEAKER_00: You know,
27:21
SPEAKER_00: when you think about a computer vision model on the back of a garbage truck
27:25
SPEAKER_00: that can identify and alert those versus a regular schedule of that.
27:28
SPEAKER_00: So I think those are the kinds of ideas we have in mind.
27:31
SPEAKER_02: Can we focus a little bit on the issue of trust?
27:35
SPEAKER_02: And you mentioned that before.
27:37
SPEAKER_02: And Stephen, I would love your thoughts on this too.
27:40
SPEAKER_02: But when you talk about trust, what exactly are you referring to?
27:44
SPEAKER_02: And what is the trust addressing?
27:47
SPEAKER_02: Where is there a trust deficit that would require you to address the trust issue?
27:52
SPEAKER_00: You know, this is a moment in America where AI is not trusted.
27:56
SPEAKER_00: And I think it's not trusted because people look at it and say,
28:00
SPEAKER_00: what data does it have access to?
28:02
SPEAKER_00: Do I have visibility on what of my personal data it has access to?
28:06
SPEAKER_00: What is it building about everything about me?
28:09
SPEAKER_00: And certainly the critical services that people rely on each and every day.
28:13
SPEAKER_00: So when we talk about one sector that we care deeply about,
28:17
SPEAKER_00: which is rural hospitals,
28:19
SPEAKER_00: because a cyber attack that disrupts those operations can impact lives,
28:23
SPEAKER_00: we also have to recognize the flip side of that is health information is some of
28:26
SPEAKER_00: the most private, sensitive information people care about.
28:31
SPEAKER_00: So we intend to invest real time in engaging local communities and coming in
28:38
SPEAKER_00: as partners with the technical expertise and the funding and resources.
28:42
SPEAKER_00: But fundamentally, they're setting the timeline based on the trust.
28:47
SPEAKER_00: A former colleague of mine from U.S. Cyber Command is now the head of Texas Cyber Command.
28:52
SPEAKER_00: He's doing a lot of work focusing on critical infrastructure.
28:55
SPEAKER_00: And we started a conversation because actually a fair percentage of
28:58
SPEAKER_00: the nation's rural hospitals are in Texas.
29:01
SPEAKER_00: So we started a conversation, again,
29:03
SPEAKER_00: of leveraging the trust they've built in local communities in Texas and coming
29:08
SPEAKER_00: in really as key partners,
29:09
SPEAKER_00: fully cognizant and aware that there is a gap in trust today for good reason.
29:14
SPEAKER_00: And we have to factor that into the work we do.
29:16
SPEAKER_01: Well, the trust issue is a huge one in so many different dimensions.
29:21
SPEAKER_01: And it is actually a circular problem in the sense that folks don't trust their governments.
29:28
SPEAKER_01: They trust their local government a little bit more than their national government,
29:30
SPEAKER_01: but they don't trust their government.
29:33
SPEAKER_01: And that whole equation has been broken a little bit.
29:36
SPEAKER_01: And many local governments don't trust their federal governments.
29:39
SPEAKER_01: You have this kind of massive trust deficit.
29:43
SPEAKER_01: Well, AI promises...
29:46
SPEAKER_01: To improve trust in the sense that government can be more responsive.
29:51
SPEAKER_01: So people see their government being more responsive, have more trust in it.
29:54
SPEAKER_01: But until government's more responsive,
29:57
SPEAKER_01: people are pretty wary about their governments using AI at all, right?
30:02
SPEAKER_01: So how do we kind of adjudicate this tension that we have where we need to use
30:10
SPEAKER_01: to be better at governing and therefore develop trust,
30:14
SPEAKER_01: but the public doesn't really trust?
30:15
SPEAKER_01: Trust government to use those tools, right?
30:18
SPEAKER_01: And so, you know,
30:19
SPEAKER_01: Ann referenced some of these things,
30:21
SPEAKER_01: and they're not just the domain of cyber issues, but just AI generally.
30:27
SPEAKER_01: So I just would put all these things together and say,
30:31
SPEAKER_01: trust is going to require transparency and openness about what the policies are.
30:36
SPEAKER_01: And the same way that governments,
30:38
SPEAKER_01: locals and state and hospitals,
30:41
SPEAKER_01: public hospitals need to get better With regulating risk,
30:45
SPEAKER_01: they need to get better at transparency about how they're using AI
30:50
SPEAKER_01: and have informed policies.
30:51
SPEAKER_01: No one of these issues is simple.
30:54
SPEAKER_01: No one is binary.
30:56
SPEAKER_01: They require careful regulation, lots of transparency, and lots of explanation.
31:02
SPEAKER_01: So we have a long way to go on trust, and it's going to be a very difficult tightrope to walk.
31:09
SPEAKER_02: For both of you,
31:10
SPEAKER_02: a lot of the data that the public institutions deal with is pretty sensitive data.
31:18
SPEAKER_02: And how are you thinking about implementing some of these tools with those kinds
31:23
SPEAKER_02: of concerns in mind?
31:24
SPEAKER_02: This sort of builds on the trust issue and also on some of the data issues,
31:28
SPEAKER_02: Stephen, that you just mentioned.
31:30
SPEAKER_01: The answer to this question is more complex than it would seem at first.
31:34
SPEAKER_01: So the easy thing to say is...
31:38
SPEAKER_01: There'll be no sharing of PII information, right?
31:41
SPEAKER_01: We're going to protect it, and that should be the goal.
31:45
SPEAKER_01: But that's complicated because can a bad actor re-identify the data through using AI tools?
31:55
SPEAKER_01: Or I just last week was in a conversation,
31:59
SPEAKER_01: very difficult conversation,
32:01
SPEAKER_01: with cities that wanted to connect air monitors,
32:05
SPEAKER_01: IoT device air monitors,
32:07
SPEAKER_01: in order to reduce the public health risk to children in under-resourced communities
32:14
SPEAKER_01: who had disproportionate amounts of childhood asthma.
32:18
SPEAKER_01: So where was PM2.5 creating childhood asthma or carbon emissions from vehicles?
32:26
SPEAKER_01: Well, that's all great,
32:27
SPEAKER_01: except even if the public health authorities and the hospitals,
32:32
SPEAKER_01: even if they remove the PII data,
32:35
SPEAKER_01: still don't really want to share The concentrations like the heat maps
32:40
SPEAKER_01: of childhood asthma for fear that they could be re-identified.
32:43
SPEAKER_01: I only go through that kind of convoluted example,
32:46
SPEAKER_01: which is to say, sure, everyone can agree on the rules broadly.
32:50
SPEAKER_01: The application of the rules becomes more complicated.
32:54
SPEAKER_01: What data is stripped out?
32:56
SPEAKER_01: What data is left in for usable purposes?
32:58
SPEAKER_01: And we see this on, you know, the flat camera debate.
33:01
SPEAKER_01: We see this on LPRs, license plate readers for parking purposes.
33:08
SPEAKER_01: What do you keep?
33:09
SPEAKER_01: What becomes in the public realm?
33:10
SPEAKER_01: What's a private actor keep?
33:11
SPEAKER_01: What's a private actor need to curate?
33:13
SPEAKER_01: Very difficult issues.
33:15
SPEAKER_01: I think that they need to be addressed,
33:17
SPEAKER_01: and I think they can be addressed, but not without a fair amount of attention.
33:20
SPEAKER_01: It's the same as this whole project that Ann and I are in together, right?
33:24
SPEAKER_01: They need kind of sophisticated attention to the rules and the application of
33:29
SPEAKER_01: the rules and a willingness to kind of spread that learning among all size cities
33:34
SPEAKER_01: and hospitals.
33:36
SPEAKER_00: You know, a couple of thoughts.
33:37
SPEAKER_00: One is I recall a conversation with the Las Vegas Police Department.
33:41
SPEAKER_00: You know,
33:41
SPEAKER_00: to your point,
33:42
SPEAKER_00: they had deployed drones distributed on roofs in Las Vegas to fight crime.
33:48
SPEAKER_00: And it had a measurable impact in the reduction of crime.
33:53
SPEAKER_00: They shared a few things that they did.
33:54
SPEAKER_00: So our question was exactly the one you asked,
33:56
SPEAKER_00: which is,
33:58
SPEAKER_00: how did you think about gaining citizen trust
34:01
SPEAKER_00: and managing concerns regarding what data is collected, etc.?
34:05
SPEAKER_00: ?
34:06
SPEAKER_00: And essentially their answer was we began with the end in mind,
34:09
SPEAKER_00: which the end was citizens who trusted what we were doing as well as a real impact
34:15
SPEAKER_00: in driving down crime.
34:17
SPEAKER_00: So they did things,
34:18
SPEAKER_00: first of all,
34:19
SPEAKER_00: in terms of communicating to citizens what they were doing,
34:21
SPEAKER_00: but also,
34:22
SPEAKER_00: you know,
34:22
SPEAKER_00: the way the drones are designed is when they're alerted to a crime incident,
34:27
SPEAKER_00: the drones turn their cameras up,
34:29
SPEAKER_00: facing upwards, rather than collecting until they're directly over the crime scene.
34:34
SPEAKER_00: So to Stephen's point about data,
34:36
SPEAKER_00: that's a really thoughtful way to say we're not even collecting it till we're over
34:40
SPEAKER_00: the crime scene when clearly it's needed to see is there a car or a person fleeing
34:46
SPEAKER_00: the scene.
34:47
SPEAKER_00: But that's one way to tell citizens, unless,
34:50
SPEAKER_00: you know,
34:50
SPEAKER_00: this is how we've thought from the beginning about managing both the collection
34:54
SPEAKER_00: and retention of data that's not necessary to the actual mission outcome.
35:00
SPEAKER_02: And this is where I think the point you raised about transparency is so important.
35:03
SPEAKER_02: How many people who are engaged in this debate know that fact?
35:06
SPEAKER_02: But that fact would be materially important for people to trust that
35:10
SPEAKER_02: the system isn't being abused, at least in this one instance.
35:14
SPEAKER_02: So it's a really great, great example.
35:17
SPEAKER_00: Exactly.
35:17
SPEAKER_00: And as documented as this is the standard of the way it operates.
35:21
SPEAKER_00: Here's the way we're transparent with the public that you can audit that this is the way.
35:25
SPEAKER_00: You know, that clearly also generates trust and it's been considered from the outset.
35:31
SPEAKER_02: If we check back with you in,
35:32
SPEAKER_02: let's say,
35:32
SPEAKER_02: four to six months,
35:34
SPEAKER_02: what do you hope and expect that we'll see as results of all your hard work
35:39
SPEAKER_02: and good work in this area?
35:41
SPEAKER_01: I think across the country in public service jobs are really good people.
35:46
SPEAKER_01: That's what I've kind of devoted my life to helping them do a better job.
35:50
SPEAKER_01: They're in local hospitals, in their local cities, in states, in government, in various places.
35:57
SPEAKER_01: I think what I would hope that they would,
36:00
SPEAKER_01: where I would hope we would be in several months is
36:03
SPEAKER_01: that we would make them more aware of what they can do to reduce the risk.
36:07
SPEAKER_01: We make them more aware of what partnerships they need in order to reduce the risk.
36:12
SPEAKER_01: We would help them understand that AI is great in terms of potential,
36:17
SPEAKER_01: but AI has risks and that there are ways to That we can help them understand
36:22
SPEAKER_01: and apply the lessons to improve the benefits and reduce the risks.
36:26
SPEAKER_00: I love the way Stephen put that.
36:27
SPEAKER_00: So I would just add a little piece to that,
36:30
SPEAKER_00: which is to say there are parts of America that often don't get the benefits
36:36
SPEAKER_00: of life.
36:37
SPEAKER_00: Advanced tech, like artificial intelligence.
36:39
SPEAKER_00: As Steve said,
36:40
SPEAKER_00: it brings real benefits,
36:43
SPEAKER_00: like knowing that,
36:45
SPEAKER_00: you know,
36:45
SPEAKER_00: city leaders can sleep well at night,
36:47
SPEAKER_00: knowing a school system won't get hacked,
36:49
SPEAKER_00: children's records stolen,
36:50
SPEAKER_00: knowing a local hospital won't be disrupted, that peace of mind.
36:54
SPEAKER_00: Great.
36:54
SPEAKER_00: And Stephen, thank you for this wonderful conversation and for joining us today.
37:24
SPEAKER_00: Thanks so much.
37:25
[music]