The RSAC annual cybersecurity & technology conference in San Francisco represents a unique opportunity to measure the concerns of the global cybersecurity community and what it is concerned about. As such, the central theme at RSAC 2019 was clear; everyone wants to know how to use artificial intelligence (AI) and protect themselves from AI. In conjunction with Unsurfaced.com's monthly Reporter's Notebook video series, Unsurfaced.com senior reporter Eric Geller interviewed our news director Michael Smith and TechTarget SearchSecurity senior site editor Alissa Irei regarding the top trends seen on stage and in the halls at RSAC 2019.
AI was prominent throughout RSAC conferences in recent years. However, as businesses' AI considerations matured, discussions at RSAC 2019 focused on the trade-offs of agentic AI solutions, the regulatory environment in the U.S. and abroad, and the future of work in an increasingly automated world. Some executive-level security professionals believe that AI can greatly accelerate and enhance the performance of their teams' tasks due to the numerous repetitive processes required in running a security operations center. AI agents can analyze and generate reports from reams of data without ever getting tired, freeing up human employees to perform more complex and nuanced tasks. Research shows companies are racing ahead with testing and implementing AI agents in their workflows.
However, AI presents many serious cybersecurity risks. The technology is helping automate ransomware attacks, accelerate lateral movement through networks, and identify targets' vulnerabilities. AI tools have flaws that can expose businesses to intrusions, and AI agents can be hijacked, leading to theft of sensitive information or disruptions to a company's normal functions.
Similarly, AI's impact on the workforce remains uncertain. Some corporate leaders are optimistic about shrinking salary budgets, while other executive-level professionals are having difficulty imagining AI agents replacing humans on complex projects, and still others are struggling to hire experienced workers in today's competitive job market.
Another theme was prevalent throughout this year's RSAC conference: the absence of the U.S. government. For political reasons, the Trump administration blocked several agencies from attending the conference, including the Cybersecurity and Infrastructure Security Agency, whose officials were unable to attend anyway because of the partial government shutdown. The Trump administration's abrupt withdrawal from RSAC highlights its trend of under-investing in cybersecurity, which has created turmoil between federal agencies and their domestic and international partners.
Michael Smith: Hello, I'm Michael Smith.
Alissa Irei: Hi, I'm Alissa Irei from SearchSecurity.
Eric Geller: And I am Eric Geller.
Michael Smith: And we are here to discuss RSAC Conference 2019, which took place last week. You guys were both present at the show in San Francisco. I was covering it from afar and wanted your thoughts on the show. What did you hear? How did it compare to the message of the show/theme heading into San Francisco last week? Alissa, why don't you go first?
Alissa Irei: Okay. So, the theme of the show was community. And I found that an interesting, pointed choice. And clearly everybody's mind is on AI. Therefore I find that the decision to emphasize the importance of community is an intentional choice. The importance of human operators. Either human "in-the-loop" or human "on-the-loop." And I think there is anxiety across various fields, not just ours, about job replacement and using AI. Therefore, the organizers of the conference, at least, were making the point that we still need humans. Artificial intelligence is not intelligent without human operators. And for our own safety, as well as the safety of others, we do need humans involved in these processes. Eric, what was your impression of the conference on the ground vs. the theme?
Eric Geller: Well, obviously, everywhere you look there was that focus on AI. Particularly looking for understanding of the threat landscape but also trying to get ahead of it with new defensive solutions. That was the common theme in a lot of sessions even if they weren't billed as AI talks. However, for my money, I think that the big theme I was noticing was the tagline on all the posters did say, as you said, "the power of community," but there was a major part of the community missing, which was the federal government, who pulled out of the show just weeks before it started, along with several other agencies. And clearly every year there are a lot of people from government who come, listen to what the community is saying, and discuss their own plans. This is one of the places where those conversations are most fruitful, according to a lot of folks I spoke to both before and during the show. There is some anxiety about what it means that the government may no longer want to participate in events like this one as much as it used to. Clearly there have been a lot of cuts at agencies that deal with working closest to the business community and researchers in the field, who make up a large amount of attendance at both RSAC and a great number of other conferences that we cover. So I think that was a striking contradiction that a lot of folks saw.
On one hand, emphasis on community. On the other hand, a major portion of the community choosing not to participate due to reasons unrelated to the show itself. Now, that will probably not happen again. I think we'll see them back at future shows, potentially even back at RSAC next year. At least in this case, you have a lot of people wondering if it sends a broader signal. And of course, we're looking for more information right now from the government about their cybersecurity strategy that they recently put out.
A lot of folks said to me, RSAC would have been a perfect place to roll out information on what that means in practice. And of course, that didn't happen. So, there were a lot of hallway conversations about the fact that there is this big void being left by some of the federal agencies that normally would be participating and even stewarding some of these conversations.
Michael Smith: Yeah, that is interesting. I know that my colleague Becky Bracken had a story about how other governments and nations brought their cybersecurity experts over to discuss some things going on in their neck of the woods, but definitely a gap was noticed.
I wrote an article a few weeks ago regarding spyware policy, sort of a potential shift in spyware policy here in the U.S., and how many of the spyware opponents working for civil society organizations and cybersecurity researchers/vendors specializing in this type of work were very fearful that a change was taking place in terms of spyware policy within the U.S. government, and how they felt there was a lack of communication. There was a lack of people who were still in government working on this and communicating: "Okay, here is our strategy. Here is what is going on. Here is where we are headed." And they were, one person told me, just sort of flying blind, that there wasn't any sort of cooperation or communication with the government at this time and they were just lost at sea. A lot of people had left their positions in different agencies and so they are just sort of winging it now, hoping for the best but really not sure.
So, Eric, to your point, I think it has made a significant difference.
Alissa Irei: It is an interesting moment. I feel like it's a moment of sort of unprecedented change, and to state the obvious, unprecedented change, and it's a moment that in a perfect world would see a lot of public-private partnerships and cooperation and input from the private sector on public regulations and legislation. So it is a notable absence, and I think one that's likely to heighten anyone's concerns about AI regardless of what the federal government is or isn't doing.
Michael Smith: Yeah, yeah, my anxiety. I'll just tell you guys what it's like up here. It's actually off camera, so it's like way above my...
Alissa Irei: Out of frame.
Michael Smith: Yeah, it's out of frame. Let's talk about AI. I know from just managing all of the stories coming in and looking at all the sessions and covering my own sessions that AI was pretty much every other session, or had some type of AI component to them, or were solely focused on AI. Obviously a big focus at the show.
The one thing I thought was pretty interesting from my perspective as an outsider, just talking to people leading up to the show and then also people that were there last week, was that there seemed to be a bit of a split, maybe not a bit; maybe that is being too gentle and too kind, but a split between C-level executives and higher-ups and what they thought about AI, and what researchers were seeing at ground level. You had a lot of researchers saying we need more human oversight. We need to be careful with agentic AI rollouts. We need to be careful with vibe coding/coding assistance, etc. We need more guardrails. We need more oversight. And then you had a lot of people, one person specifically, whose name escapes me, but he spoke at one session and said that human oversight: we need to get rid of it because it will slow things down. That is exactly why we want to use AI, to speed things up! What did you guys see or hear there?
Alissa Irei: It appears as though on the business side, there's such excitement about developing new uses for AI and experimenting with new AI technologies, and "asking for forgiveness" rather than seeking "permission." From what I saw and heard at the conference, I believe that will certainly provide opportunities for bad things to occur. I agree with Eric that he has written about the vulnerabilities introduced by vibe coding and the lack of oversight. As such, I view the situation as quite troublesome, at minimum. Conversely, I agree with your point, Michael, regarding the C-level. I attended a session in which the chief information security officer (CISO) of Exabeam discussed how the agentic AI technology they've deployed in their security operations center (SOC) discovered a North Korean malicious insider whom they had employed. The individual was hired as an employee and the agentic AI identified the insider's malicious behavior within hours, if not minutes, after the insider logged onto his computer account for the first time. I perceive that as an encouraging example of AI functioning effectively. However, I'm unsure how consistently effective the technology is, and I fear that managing the enormous vulnerabilities introduced by AI could be alarming to many individuals.
Eric Geller: That quote you mentioned really stood out to me during that panel I reported on. He essentially stated, "If AI developed your YARA rules, you should delete them immediately because they're likely trash." I think that quote illustrates this hunger for automation, this desire to simply make things easier for businesses, and this desire for higher profit margins. When you automate or outsource your employees' jobs, you increase your profit margins. Your company will appear more successful to investors. Venture capital firms will provide more funding for your company. I believe that AI represents only a small portion of this issue. The majority of this issue revolves around appearing profitable while reducing the amount of labor required to achieve profitability.
I think one of the themes evident throughout many of the discussions involving AI was not a matter of finding balance. Rather, it seemed to be both simultaneously. Yes, you may wish to implement some type of agentic solution to take those mundane tasks away from your specialized expert humans. However, you will also require some form of governance framework to ensure that a human will periodically monitor and review what your AI agents are accomplishing and what privileges they are receiving. If you employ an AI agent that becomes uncontrollable, you'll realize the problems associated with its malfunction when you periodically check in to inspect what it's doing. For instance, if the agent is incorrectly identifying or labeling items improperly, you will observe evidence of that. Therefore, I believe that is where many of the discussions concluded: yes, there are valid reasons why corporate executives in particular are searching for methods to modify the function of their analysts and to introduce more AI into the threat analysis aspects of their jobs. However, similar to how you will need human supervisors to oversee human employees, you will need human supervisors to oversee your AI agents, because neither humans nor machines are flawless. Furthermore, considering the scale at which some of these corporations operate and the financial implications of either securing or failing to secure their networks, we are discussing significant amounts of money that can be earned or lost. As such, you will want a human reviewing the output produced by your AI agent.
Michael Smith: Yeah. I understand what you're saying. The presentation that I covered last week was presented by Check Point researchers and they basically stated that the security measures developed over the past twenty years to protect networks, improve endpoint defenses, and move execution into the cloud, where it is presumably safer, have in practice many times been punched through by these AI coding assistance tools, setting security back ten years. They literally stated that security has regressed ten years due to these tools providing attackers a path from an employee's endpoint to their most sensitive information via their development environment. All of this work completed over the previous 10 to 20 years is now, in their view, irrelevant. Additionally, I believe one of the factors that shocked them was how many companies rushed to these tools without realizing that regardless of whether or not they contained a critical flaw or exploited a weakness, they were creating a tunnel from a typical workstation, that is probably still inadequately protected today, to areas of their network containing high levels of privilege. They were clearly taken aback by people rushing headlong toward these products without pausing long enough to consider whether this was a wise decision, whether they should apply additional protections to safeguard the products they provided to users, and whether they should establish additional controls to monitor what their agents were doing with respect to permissions granted via vibe coding tools. Based on my observations at the show and comments from attendees, I am uncertain whether this mentality will shift anytime in the near future, despite numerous studies indicating numerous potential vulnerabilities and increased threat surfaces resulting from AI.
It seems to me that instead of slowing down with regards to using AI for automated processes, that is, utilizing AI for cost reduction, organizations continue to feel increasing pressure to maximize their investments in AI. As Eric noted previously, the goal is to shed costs, save money, and reduce workforces.
Michael Smith: And that makes perfect sense to me. I recall writing a story after attending a Check Point session last week. Essentially, Check Point researchers indicated that they spent 20 years building up all types of security measures to defend against network attacks, strengthen endpoint defenses, and transition execution into the cloud where it is supposedly safer. These AI coding assistants were able to punch holes through all of these security measures and reset security ten years backward. The researchers essentially stated that it now provides attackers with a route from employees' endpoints directly into development environments, which did not use to be possible. Thus, all of this work performed over the previous 10 to 20 years has become irrelevant.
In essence, one factor that caused them concern was the fact that many organizations rushed headlong toward these tools without acknowledging that even if no flaws existed in these tools, they were still establishing a tunnel from an unsecured employee workstation to potentially very sensitive areas of their network containing high levels of privilege. They expressed great surprise that many organizations approached these solutions with little caution or consideration regarding how safe or sensible it would be to utilize them.
And personally, I was surprised by their expressions of surprise, and by my own surprise when hearing their concerns about such attitudes among organizations relative to adopting these solutions. Similarly, based on my experiences attending this show and listening to presentations, I am uncertain whether such mindsets will change anytime in the near term, despite multiple analyses demonstrating various vulnerabilities and risks associated with AI as well as expanding threat surfaces attributable to AI.
Michael Smith: Oh yeah... the power of community...
Alissa Irei: Actually, I think Eric's comment reminded me of an impromptu conversation I had with Diana Kelly, who is the CISO at Nomad Security. She delivered a presentation on model collapse and the eventual inevitability, and possibly a reference to the movie Idiocracy, that eventually we'd all just become extremely foolish if these models continued consuming their own content. Which relates back to another theme regarding community and the importance of human input, contributions, and intelligence.
Additionally, I'll attempt to remain optimistic. There were several instances at this conference, including Exabeam's CISO's presentation that I referenced earlier, where I perceived some positive examples of AI operating as intended in the SOC. And as we know, SOC analysts are overwhelmed and stressed out, and if AI agents can assist them in alleviating some of this workload, sifting through noise and bringing forward actionable elements, that would be fantastic.
