Meet with Cobalt at Infosec World 2026

The security industry is drowning in AI claims. Every vendor promises autonomous everything, but alert volume keeps rising and validation gaps keep widening. Teams are left sorting signals from noise, with fewer hours and fewer people to do it.

Point-in-time testing and script-based scanning cannot keep pace with modern release cycles, expanding attack surfaces, and AI-driven threats. Tools alone cannot tell you whether a vulnerability is exploitable in your environment. Only experience can.

From Adoption to Exposure: What 2026 Pentesting Data Says About AI Risk

Speaker: Gunter Ollmann, CTO, Cobalt
Date & Time: Tuesday, October 13 · 10:30–10:45 a.m. · Tech Theater 1

The gap between AI adoption and AI security is widening. Most organizations are integrating LLMs faster than they are recalibrating their pentesting strategies to account for how these tools change the attack surface. Gunter Ollmann breaks down the latest research on how leading security teams are, and aren't, adapting to machine-speed threats.

Catch the talk, then keep the conversation going. Stop by Booth #608, or let us buy you a coffee and talk through what the data means for your environment.

AI-powered offensive security testing grounded in human expertise

Humans over hype is not a tagline. It's the architecture. The Cobalt Offensive Security Platform runs AI-driven discovery and orchestration at scale, governed by the Cobalt Core, a global network of 500+ vetted pentesters. The Cobalt Core validates exploitability, uncovers business logic flaws and chained attack paths automation misses, and prevents autonomous testing from introducing unintended risk into your environment.

At Booth #608, talk methodology with the people who run it: Cobalt executives and solutions architects, on the record about what the platform does, how findings are validated, and where automation stops and human judgment starts.

Key Benefits:

Go deeper on Cobalt research. Bring your questions from the Tech Theater session directly to a solutions architect at Booth #608.
Talk to the people who build the strategy. Get direct answers from Cobalt executives on where the industry is headed and how continuous offensive security testing fits your risk posture.
Design your program with an architect, not a script. Work through your specific environment and identify where exposure windows exist today.
See the Cobalt Platform in action. Learn how Cobalt launches a pentest in hours and delivers real-time findings, not a quarterly report.
Access data you cannot get elsewhere. Learn how our dataset of 5,000+ annual pentests and 10,000+ critical or high-severity findings benchmarks against your industry.

While many continuous solutions rely solely on AI and scripts, the human validation provided at Cobalt is the key differentiator. By leveraging Cobalt's pentesting expertise, we move beyond the noise of raw data, allowing our team to focus on high-impact remediation rather than manual de-duplication.

Jon Cheuvront

Sr. Security Engineer, Gallagher

DEF CON 34

Hacking Big Iron: When Modern Security Assumptions Fail on Mainframes

DATE TBA

Mainframes still underpin critical infrastructure such as banking, airlines, and government systems, yet most modern security teams approach them using assumptions formed around Unix, Windows, and enterprise platforms. These assumptions often fail on z/OS, creating blind spots that are difficult to detect and easy to underestimate.

This talk explains how mainframe security actually works and why familiar concepts such as "root," shells, ports, and lateral movement do not translate cleanly. Focusing on components like JES, JCL, RACF, CICS, and PR/SM, we explore where attackers and defenders truly operate today: transactions, security managers, and management boundaries.

BLACK HAT USA

MCParasite: Universal MCP Worm Security Testing Framework

Aug 6, 2026 | 11:20 AM PDT

MCParasite takes prompt injection and tool poisoning techniques that already exist and chains them into a self-propagating worm that lives inside MCP. A single rogue server poisons the agent's context the moment it connects.

The agent then writes a payload to whatever channel it has access to: Slack, GitHub, Jira, email. A completely separate agent on a different system reads that message, gets infected, and starts executing commands on its own. Shell access, credential theft, data exfiltration. The victim agent never touched the malicious server. We tested 8 models from 4 vendors. 5 out of 6 propagated the worm. We are actively expanding to DeepSeek and others; the talk will cover full results across all major frontier models.

DEF CON 34

Deep Dive into Fuzzing (Course)

Aug 10 - 11, 2026 | 8:00 AM - 5:00 PM PDT

Attendees would be emulating techniques which would provide a comprehensive understanding of "Crash, Detect & Triage" of fuzzed binaries or software. In "Deep dive into fuzzing" we will be covering a detailed overview of fuzzing and how it can be beneficial to professionals in uncovering security vulnerabilities with a hands-on approach through focus on labs.

Finding vulnerabilities in software requires in-depth knowledge of different technology stacks. Modern day software’s have a huge codebase and may contain vulnerabilities, manually verifying such vulnerabilities is a tedious task and may not be possible in all cases. This training is designed in such a way that it introduces the concept of fuzzing and vulnerability discovery in software’s covering multiple platforms such as Linux & Windows and triage analysis for those vulnerabilities.

INFOSEC 2026 - ORLANDO, FL