Can you crack the code? Take on the TrendAI™ Vault Challenge for your chance to win $10,000 USD. Collect code digits by engaging with TrendAI™ experiences throughout Black Hat and unlock your opportunity to claim the prize.

*Open to US residents only, subject to US taxes

Watch proactive exposure management, AI security, and threat detection and response merge into a single workflow that clearly shows the use case for consolidation as concrete, not theoretical.

Speak directly with the researchers behind the disclosures shaping next year's risk conversations with your board. Learn about the latest findings from TrendAI™, including details from an AI-powered sweep of nearly 19,000 MCP server projects cloned from GitHub on Wednesday, August 5 at 12:30PM.

On Thursday, August 6th at 2pm, you’ll have the opportunity to hear from industry expert Kevin Poulsen as he shares how he went from the FBI's Most Wanted list to become an award-winning investigative journalist and the author of Kingpin, the story of real-life computer hacker Max Butler, who, under the alias Iceman, accessed over 1.8 million credit card accounts.

The AI application you'll defend doesn't just answer questions. It acts. It reads support tickets, pulls up customer accounts, issues refunds, and resolves them on its own, and that autonomy is what makes it a new kind of attack surface. You'll start with a completely unprotected agent and watch the attacks land, then turn on Trend Vision One features one at a time and see how each one hardens it.

First the model. You'll trick a real Amazon Bedrock model with a direct prompt injection, then with one hidden inside an uploaded document. Turn on AI Guard and both are blocked at the input. Then slip a malicious file into the app's S3 storage, turn on File Security, and watch it get caught and quarantined before it reaches the model.

Then the real test: the agent itself. Unprotected, a hidden instruction buried in an ordinary support ticket hijacks it into leaking other customers' data through its own reply, an action it's allowed to take, which is why permissions alone don't stop it. Turn on AI Guard around the agent and run the same attack. This time Vision One inspects what the agent reads and what it sends back, and stops the data before it leaves. You'll finish in Trend Vision One, where every attack you ran shows up as correlated detections in one place.

TrendAI™ will take you inside an AI-powered sweep of nearly 19,000 MCP server projects cloned from GitHub. The initial scan surfaced more than 17,000 potential vulnerabilities, and after multiple rounds of validation, hundreds of genuinely exploitable issues remained, including SQL injection, remote code execution, and path traversal. The findings reveal a fast-expanding MCP attack surface and a harder truth for security teams: LLM-powered analysis can surface real risk at scale, but without expert verification, it can also flood teams with confident noise.

The traditional linear Cyber Kill Chain is dead, obliterated by AI-driven automation that converts a step-by-step progression into a continuous, hyper-velocity loop. Defensive strategies built on breaking a single static link cannot survive an adversary that rewrites its playbook mid-execution. Neutralizing this relentless cycle requires a radical shift toward real-time vulnerability shielding, backed by ground-level human threat intelligence capable of decoding nation state adversarial modus operandi and collaboration among cybersecurity companies and governmental agencies. 

Celebrate the people driving cybersecurity forward. Hosted by the TrendAI™ ZDI, the Vanguard Awards recognize the researchers, vendors, and industry leaders making an impact through vulnerability research, responsible disclosure, and cybersecurity innovation.

Join us as we acknowledge the achievements of the security community and celebrate another year of advancing cybersecurity together.

The session aims to explain the critical importance of AI security beyond mere vulnerability management. Our findings showcase security gaps and vulnerabilities in the MCP server deployments. Attendees will learn how a vulnerable or misconfigured MCP server can lead to a serious breach of the company’s infrastructure, and where to focus their attention on the secure implementation of MCP servers inside a corporate environment. The session will also highlight best practices for MCP risk mitigation based on security research of 19,000+ open-sourced MCP servers.