We walked into Black Hat with a lot of curiosity around pentesting, runtime defense, SOC automation, and agent security. What we found: the gap between vendor marketing and real security challenges is narrowing, but it’s still early days for most of what’s getting airtime.

To kick off the week, we hosted an invite-only breakfast for senior security leaders in the La Luna room at the Four Seasons. It was a quiet, grounded way to start the day, with strong coffee, better food, and honest conversations before the chaos of the show floor began.

Here’s what stood out, and what we’re still thinking about:

Explosion of Agentic Pentesting Companies

Pentesting may be entering its most transformative phase in a decade. Live demos showcasing real-time exploitable findings stood out against the static, PDF-heavy approach that has dominated the industry.

The Startup Pavilion reinforced the momentum: new players offering continuous, AI-powered offensive testing drew consistent traffic, validating the appetite for something beyond quarterly engagements.

From practitioner conversations, the shift is clear: security leaders want real-time adversarial intelligence integrated into CI/CD pipelines, rather than relying on static testing cycles.

Runtime Security and Cloud Application Detection & Response (CADR)

Runtime security made its mark, especially in the area of CADR (Cloud Application Detection & Response). Both corp dev teams and practitioners saw value in runtime scanning to eliminate false positives and identify true reachability, a promise that is now feasible thanks to clever implementations of eBPF.

The broader signal from Black Hat: runtime isn’t just additive anymore. It is becoming the operating truth for defense. The shift reflects a move from visibility to protective response, especially as cloud-native environments demand faster detection-to-action loops.

Vulnerability Management Stays Top of Mind

Even with all the buzz around AI, vulnerability management was everywhere and very top of mind for CISOs.

But it’s no longer about finding more flaws. The shift is toward continuous exposure management, blending scan results with exploitability and business impact across all assets.

The new mandate is to prioritize and remediate at scale.

This is a category that isn’t going away anytime soon. But the battleground is shifting toward continuous exposure and automated remediation. With increasing overlaps with the runtime approach, the question now is whether once-distinct players will begin to converge.

AI SOC: Crowded but a Real Need

The AI SOC conversation continues to be one of noise and clarity at once.

With dozens of players taking variable approaches, we leaned on our practitioner network to distill some truths:

• Analysts aren’t going away: AI is best positioned for triage and noise reduction, not high-stakes decision-making.

• Context is non-negotiable: Without rich environmental context, analyst trust evaporates.

• Implementation is messy: Poor data quality kills AI value; structured telemetry and workflow hygiene must come first.

• New models are emerging, ranging from triage-only plays to hyper-automation and agentic SOCs that handle investigations end-to-end. While triage is the easiest to experiment with, practitioners are excited about more holistic approaches that augment them.

The space is saturated, but it’s not a mirage. The winners will be those who move beyond alert reduction into investigation depth and autonomous workflows.

Surprisingly Quiet AppSec Booths

In sharp contrast, many of the larger AppSec booths felt quiet. That silence may reflect two things:

1. Buyers see legacy AppSec as “good enough” in the current cycle.

2. AI-generated code is exposing gaps that traditional tools weren’t built to address. 

There was also real interest in early researchdemonstarting how automated vulnerability discovery and delta analysis could transform CI/CD pipelines, hinting at a very different AppSec future.

AppSec could be in a lull before reinvention. The next big wave may be AI-native platforms that continuously detect, fix, and govern AI-generated code in real time.

Securing AI Agents

Some of the most forward-looking conversations were about a category that is just starting to emerge: AI Agent Security.

Across talks with vendors and CISOs, we kept hearing the same concerns:

• Identity creep — agents accumulate excessive permissions over time

• Observability gaps — agents generate far more logs than humans, straining SOC pipelines

• Agent-to-agent risks — early fears about opaque communication channels highlight oversight challenges

• Workflow security — as enterprises deploy multi-node agents (with Slack and email access, even meeting attendance), securing their actions becomes critical

It feels a lot like container security in 2014: nascent, fragmented, but inevitable. Expect identity, data, and observability solutions to pivot quickly toward agent security.

Final Takeaways

Black Hat 2025 made one thing clear: security teams are no longer satisfied with passive visibility or one-off testing cycles. From agentic pentesting to runtime defense, from vulnerability management at scale to the first signs of an Agent Security stack, this year felt like a fundamental shift in mindset.

AI was everywhere, but the conversations that mattered most focused on how to make security workflows faster, more integrated, and actually actionable. Buyers want fewer dashboards, fewer PDFs, and fewer alerts. They are seeking real-time signals that can plug directly into how modern engineering teams build and ship software.

This isn’t about trends. It is about meeting the speed and complexity of modern systems with security that can actually keep up – and we’re looking to find the founders and platforms meeting the need.

This article is for informational purposes only and does not constitute investment advice. Views expressed represent the opinions of Jump Capital. Jump Capital may have investments in or pursue investments in the security technology sectors and companies discussed. References to specific companies do not constitute investment recommendations.