Interviews | July 15, 2026
The line between national security and enterprise security is blurring quickly
DropZone AI | Fortra | Pentera | Sublime Security
Q1. Security operations centers are under intense pressure to do more with smaller teams and rising alert volumes. Where do you think autonomous AI analysts like Dropzone can realistically replace human effort today, and where is human judgment still indispensable?
The question isn't whether AI can replace analysts. The more important question is how AI can help security teams work on all the projects that they never had the capacity to do.
The challenge facing most SOCs isn't simply alert volume. It's that security operations don't scale linearly with headcount. Every year organizations deploy more security controls, generate more telemetry, and face more sophisticated threats. They don't add analysts at the same rate. Eventually the gap between analytical capacity and what’s required by the business grows so big the SOC teams become perpetually overwhelmed.
That's why we think the future isn't simply using AI chatbots to help investigate security alerts. It's the Agentic SOC, where one human team is augmented by a team of specialized AI agents that can investigate alerts, conduct threat hunts, analyze threat intelligence, and work together continuously across the environment.
An investigation agent can determine whether suspicious activity warrants escalation. A threat intelligence agent can identify emerging campaigns and provide context. A threat hunting agent can proactively search for related activity. When those capabilities work together, you can automate a much larger portion of security operations than any single agent could handle on its own.
Human judgment remains essential. Security practitioners still need to make decisions about risk, business impact, priorities, and response.
What changes is the execution model. Instead of being constrained by the number of investigations, hunts, or analyses a team can perform in a day, organizations can deploy autonomous agents that operate continuously across those functions.
The most effective SOCs in the future won't be the ones with the largest teams. They'll be the ones that can bring together human expertise and machine-scale execution.
Q2. What has to happen technologically and operationally before organizations can fully trust completely autonomous security investigations? How do you build that trust with enterprise customers, particularly in regulated industries where accountability for security decisions is a significant concern?
CISOs don't get to outsource accountability to AI. If something goes wrong, the security team is still responsible. That's why trust can't be based on marketing claims or benchmark results. It has to be based on transparency, control, and consistent performance.
Security teams have spent years being asked to trust black boxes. Most of the time that hasn't gone particularly well. If an autonomous agent determines that activity is benign, suspicious, or indicative of a real security incident, organizations need to understand why. They need visibility into the evidence that was collected, the hypotheses that were tested, the findings that supported the conclusion, and the reasoning behind the recommendation.
That's why we've built Dropzone around a GlassBox AI model. Every investigation is explainable, auditable, and easy to inspect. We don't think autonomy and transparency are competing goals. In practice, we think autonomy only works when transparency exists.
Trust is built through a learning loop, not a handoff. We recommend customers start in assistive mode, feed the agent context about their environment, and run reviews between AI and analyst conclusions to improve the whole system. Automated response is something customers should selectively implement after establishing confidence in true positives.
We are not replacing human judgment; we are making it stronger with better evidence and better process. The audit trail is built in from day one. That is how you earn trust with a CISO and their regulators.
The goal is to remove humans from repetitive operational work so they can focus on governance, risk management, and strategic decision-making. That's ultimately how organizations scale security operations while maintaining accountability.
Q3. What should attendees expect from Dropzone at Black Hat USA 2026? Are there any big highlights, sessions, or experiences you're particularly excited about?
At Black Hat, we're excited to show the next phase of the Agentic SOC.
Most organizations treat investigations, threat hunting, and threat intelligence as separate activities that all compete for the same limited analyst time. We think those functions should work together through autonomous agents that can operate continuously and at scale.
One of the biggest things we'll be showcasing is our new AI Threat Hunter. The challenge with threat hunting isn't just execution. It's deciding where to focus. Security teams are constantly balancing new threat intelligence, emerging attacker techniques, and limited analyst time. Most organizations have 10x more hunting ideas than they can realistically pursue.
With AI Threat Hunter, analysts can define what they're looking for or select from hunt packs developed by Dropzone's threat research team, and the agent handles the execution. It generates hypotheses, searches across SIEM, EDR, and cloud to filter out explainable or common activities, and investigates anomalous behaviors in depth autonomously. A process that traditionally takes days of analyst effort can be completed in about an hour, with no analyst involvement during execution.
Our Threat Intelligence Agent will be in beta as well. Intelligence is only valuable if it changes what you do next. That's why we're focused on connecting intelligence directly to operations. An intelligence agent can identify an emerging campaign or attacker technique, generate a hunt package, and hand it directly to AI Threat Hunter for execution. That's when you start seeing the value of specialized agents working together.
We also look forward to participating in the Black Hat NOC. One of the things that makes the NOC special is that it moves the conversation from theory to practice. As the industry explores how AI fits into security operations, it's important to evaluate these technologies in real operational environments. We're excited to be part of that effort.
Q1. Fortra recently launched a Defense and Intelligence Unit focused on delivering cyber capabilities to allied nations and has been pursuing FedRAMP High authorization for its data classification capabilities. How is the growing overlap between national security and enterprise security changing what customers expect from cybersecurity vendors?
The line between national security and enterprise security is blurring quickly, and that’s fundamentally changing the expectations of cybersecurity vendors. Historically, enterprise customers focused on protecting systems and networks, while national security organizations were focused on mission assurance and operating in highly constrained environments.
Today, these worlds are converging. Enterprises--especially critical infrastructure and global companies--are now dealing with nation-state-level threats, operating across hybrid and multi-domain environments, and increasingly needing to share sensitive data across partners, regions, and trust boundaries.
Because of this, customers are no longer just looking for point security tools. They’re expecting vendors to help them secure, govern, and operationalize data itself. That includes capabilities like classification, policy-based access, and control across on-prem, cloud, and air-gapped environments (as well as alignment with frameworks like FedRAMP and broader government standards). They also expect solutions to be interoperable, resilient, and usable in real-world operational contexts.
What we’re seeing is a shift toward a more data-centric and mission-oriented security model, where the question is no longer just, “Is the system secure?” but “Can we trust, share, and act on this data under real-world conditions?” Vendors that can bridge that gap — bringing enterprise-grade scalability together with national security-grade control and assurance — are the ones that will be most relevant going forward.
Q2. How does email security need to evolve to keep pace with AI-enabled phishing threats? What capability do you think email security platforms will need most over the next two years?
Email security needs to move from signature-based filtering to context-aware, intent-based detection that can identify impersonation, social engineering, and abnormal behavior even when the message is technically “clean.”
As attackers use AI to create more convincing phishing and impersonation attacks, email security must evolve beyond static filters and basic indicators of compromise. The next generation protection will need to understand context, analyze sender and message intent, and detect subtle anomalies in communication patterns in real time. Over the next two years, the most important capability will be AI-powered contextual reasoning that can distinguish legitimate business activity from deceptive behavior before a user ever acts.
Q3. Which of Fortra's offensive and defensive security technologies does the company plan to highlight at Black Hat USA 2026?
At Black Hat USA 2026, we'll be highlighting a range of offensive and defensive capabilities that reflect the industry's shift toward more data-centric and intelligence-driven security. That's reflected in our unified data security platform. The Fortra Platform goes beyond discovery and classification to label, protect, and operationalize data, alongside AI-driven controls that can discover AI usage, identify risky behavior, enforce policy, and prevent sensitive data exposure in AI apps.
Complementing this is Advanced Brand Protection powered by AI and deep threat intelligence for rapid phishing mitigation and full takedown support, as well as Threat Trace, a unique program delivering threat actor attribution at scale with actionable analysis. We'll also preview Continuous Threat Exposure Management (CTEM), launching in late 2026, which enables organizations to continuously identify, validate, and prioritize exposures based on real business risk, helping drive more strategic remediation.
On the offensive and research side, we're expanding Red Team tools functionality, including in‑memory C scripting (Beacon Interpreter), improved payload and C2 flexibility, and ongoing additions of stealth-focused tools and EDR evasion techniques. Cobalt Strike Research Labs will give customers exclusive access to emerging attack methods and advanced evasion techniques, a rich knowledge base, training, and a private Slack community.
Finally, we're continuing to evolve email security with new AI/ML models for email security that analyze content, assess risk, and detect increasingly sophisticated threats like AI-enabled phishing and vishing. Together, these advancements are helping organizations take a data-centric approach to security that reduces AI risk and strengthens cybersecurity posture in increasingly complex environments.
Additionally, on Wednesday, August 5 at 3:30 p.m. on the Main Stage, Fortra will present Operators Unfiltered: AI, Red Teaming, and the Reality from the Front Lines. In this candid, no-slides discussion, red teamers from Fortra's Outflank and Mandiant/Google, will compare notes from the front lines, unpacking how AI is actually influencing offensive operations today — and where it falls short. Note: The speaker's views and opinions are their own.
Q1. Where do you see Automated Security Validation landing in the security stack over the long-term? Do you see ASV as a standalone program or do you expect it will eventually become part of broader platforms like exposure management or CTEM?
The honest answer is both, but let's start with the principle that makes validation non-negotiable regardless of category labels: you should never trust a control you haven't tested. That's not a product pitch. It's a security truth that predates frameworks and will outlast them. In the AI era, that truth gets sharper. Vulnerabilities and exploitability move faster than any team can track manually. The time between disclosure and active exploitation keeps shrinking. Validation isn't a quarterly audit activity anymore - it's operational hygiene.
Now, where does it sit structurally? Gartner's 2026 Market Guide consolidated breach-and-attack simulation, automated pentesting, and continuous red teaming into a single category: Adversarial Exposure Validation. CTEM is the operating model that frames all of this - not a product, but a system. Inside it, two layers have to work together: Exposure Assessment Platforms surface and prioritize what might be wrong. Adversarial Exposure Validation proves what's actually exploitable. One gives you a list. The other gives you evidence.
That distinction matters. Validation is the credibility engine of the entire stack. Without it, exposure management is just a prioritized guess. The hard part - safely executing real attacks in production and standing behind the results - won't commoditize. That depth is defensible. The best security teams have already stopped debating the category question. They ask one thing: can I prove my security controls work this morning? A security leader who answers that with evidence instead of a CVSS score owns the conversation. That's where Pentera planted its flag. Proof over theory. Every time.
Q2. Organizations are deploying more autonomous systems, ephemeral environments and AI-driven development practices. How will offensive security testing need to evolve to keep pace?
The half-life of an environment is collapsing. A container spun up at 9am is gone by lunch. Code shipped by a development agent at 2am is in production before anyone scopes a test. By the time a traditional pentest is scheduled, you are measuring extinct risk — a snapshot of infrastructure that no longer exists.That's the structural break. Traditional offensive testing was built for static targets. The attack surface is no longer static.
The first evolution is table stakes: testing has to run continuously, at machine speed, in parallel with the environments it covers. Call it DevOps pace, AI scale, whatever framing fits - if your validation can't keep up with deployment, it isn't validation. It's archaeology.
The second evolution is where it gets harder, and more interesting. AI-driven development doesn't just accelerate delivery. It manufactures new classes of exposure that traditional scanning was never built to find: prompt-injectable agents, over-permissioned service accounts wired into automation pipelines, secrets bleeding through model context windows, identity sprawl across workloads that vanish by morning. These aren't classic CVEs. There's no signature for them. The only way to know if they're exploitable is to act like an adversary and actually try. That's the principle that has to hold regardless of how the tooling evolves: you cannot model your way to ground truth in an environment that rewrites itself daily. You need proof.
Pentera's premise is that the same AI expanding the attack surface can drive the testing that covers it- orchestrating and adapting attack execution continuously, under strict control. Not to replace human judgment, but to give it evidence worth acting on. Offensive security wins or loses on speed and credibility. The tools are catching up. The question is whether they can deliver both at production scale.
Q3. What are Pentera's goals at Black Hat USA 2026? How do you plan on engaging with customers and other attendees at the event?
Black Hat draws the people who built the attacks, wrote the defenses, and know exactly where the gap between the two lives. Nobody here is moved by decks or demos that don't survive contact with a real environment. That's not a challenge - that's exactly the condition we want.
We'll be showing real validated attack paths. Live environments. No slideware. If it doesn't hold up under scrutiny, we don't show it. Beyond the demonstration, three things matter to us this year.
First, defining where adversarial validation is headed. We're applying AI to execute and adapt attacks safely in production - and this audience will find the edge cases faster than anyone. That's not a risk we're managing, it's the point.
Second, technical debate. The sharpest feedback we've ever gotten came from practitioners who walked up skeptical and stayed to argue. If you think our approach has a gap - in how we keep production testing safe, in how we handle AI-generated attack surfaces, in where CTEM goes from here - we want that conversation. Ten hard technical debates over a hundred badge scans, every time.
Third, we're here to listen. The frontier right now is specific: ephemeral infrastructure that won't sit still for a test, AI agents with over-permissioned identities, the gap between "we patched it" and "we can prove it's fixed." That's exactly where we're building next - and the people in this building are closest to it."
Come find us. Bring your hardest questions. We're not here to convince you - we're here to show you and let you decide.
Q1. What do you see as the optimal balance between human expertise and machine decision-making when it comes to using AI to analyze and triage email threats? What risks do you see emerging when organizations lean too heavily on either?
CISOs are being sold a future that most vendors haven't yet built. Leaders are expected to evaluate AI solutions under the impression that these tools can immediately get the job done. This overpromise of autonomous agents to combat attackers leveraging similar technologies sounds great in a product demo, but if you look a little closer, you’ll find most vendors are still very much in the early stages of delivering value.
The issue is what to ask these vendors and how to ask them to surface that information. If you are going to an agent-run environment, how do you ensure it will actually work? Most vendors will claim autonomy outright. In my opinion, the concept of full autonomy is not a SKU; it cannot be purchased. It’s a journey, built on trust between the vendor and customer. Here's why that's so hard to catch: most of these systems are black boxes.
Opacity is a security problem, not just a UX problem. When a black-box agent takes action, it can be difficult to understand why it made that decision. If you cannot see the reasoning, you cannot verify it. If you cannot verify it, you cannot trust it. And if you cannot trust it, why would you grant it autonomy?
That said, the risks run in both directions. Organizations that overcorrect find themselves losing a different race, as attacks scale faster than most security teams can handle. Trying to match this speed with humans in the loop alone leads to analyst burnout, poor judgment, and compounding coverage gaps.
In the end, I don’t think the answer is to max automation or max oversight. It is a trust progression. Start with active human approval gates and expand autonomy only as the system proves itself in your environment.
Q2. Where do you see AI creating genuinely new capabilities in email security, beyond just identifying bad messages? What use case do you think will have the biggest impact on security teams over the next two years?
I see two primary areas where AI could have the biggest impact: self-healing security posture and phishing simulations. These two areas share a common theme: closing the loop on what a security platform misses.
A self-healing security posture is one in which triage and detection engineering agents collaborate to identify and close gaps within hours or minutes, not days. No security product is 100% effective, so the KPI we should measure isn’t detection rate, it’s the time to close a gap after a miss. Historically, when an attack bypasses a platform defense, a human has to spot it, file a ticket with support, wait for a vendor update, or apply a temporary workaround by digging through settings to add an exception. Shifting from “did we catch everything?” (which is largely unanswerable) to “how fast can we adapt when something is missed?” would be a major boost for security teams.
The other area is phishing simulations. Today, organizations rely on template-based, predictable simulations, leading employees to recognize the simulation rather than real-world attack TTPs. These are generic representations of real attacks, but they often don’t reflect what an organization is actually facing. This is compounded by pass/fail metrics that are disconnected from the security platform; ultimately, many organizations end up treating employees as a liability rather than their highest-value security layer. AI can generate more realistic simulations, personalized to each user, to better reflect that employee’s attack surface. This brings adaptability and insight, allowing security teams to evolve simulations to target known weaknesses and level up employees.
Q3. What are Sublime's plans at Black Hat USA 2026? What themes and topics do you plan on highlighting at the event?
Black Hat is where we get to talk to the people who actually feel the gap between what their email security vendor promises and what shows up in the inbox. That's the conversation we're here to have.
The theme we're anchoring to this year is tailored detection. Attackers are no longer sending generic phishing campaigns. They research your industry, your org chart, your vendors, and your naming conventions. Generic detection built on a broad signal doesn't catch that. The subject lines we're showing on our booth screen are real attacks pulled from actual campaigns targeting manufacturing, real estate, and education, designed to look exactly like legitimate mail for those organizations. Most tools miss them. Sublime doesn't, because it is built to understand how your organization is specifically attacked.
Our two AI agents, ASA (Autonomous Security Analyst) and ADÉ (Autonomous Detection Engineer), are a big part of what we're showing. ASA triages and investigates suspicious email automatically, including user-reported phish, without analyst involvement. ADÉ watches for new threats and continuously adapts your detection coverage to your environment. Together, they represent what we mean by agentic email security: not a product that requires your team to configure and maintain detections, but one that does that work for you and gets more specific to your organization over time.
The outcomes we're anchoring to are real: customers seeing 20x more attacks detected, 61% faster investigation time, 80% less time spent managing email security overall. Those numbers come from customers who replaced tools they thought were working.
Attendees should come by the booth if they want to talk through how detection that actually knows your organization is different from what they're running today.
