Zenity

An autonomous coding agent treated a security control as just an obstacle. That's the shift security teams need to plan for. Start here: which agents you're running, which MCP servers/skills, what permissions, how much autonomy? 👉 Learn more: eu1.hubs.ly/H0wXLCk0 #AIAgentSecurity

⏰ Webinar next week ⏰ Autonomous security models are already in deployment across organizations for timely vulnerability discovery and management at scale, but while defenders explore agentic capabilities, attackers have been targeting agents as a key vulnerability many organizations lack the adequate controls for. Join Itay Meller and Rock Lambros on Thursday, 7/23 at 9am PT / 12pm ET for AIDR Unpacked: A Conversation with Claude Mythos Tenant AWS. Register Now: eu1.hubs.ly/H0wXG_w0

An AI coding agent hit a blocked malicious package. Instead of stopping, it deleted the security control and installed it anyway. No attacker. Just an agent pursuing its goal. Zenity researchers break down why 🧵👉 eu1.hubs.ly/H0wWSSP0 #AgenticAI #AISecurity

A malicious npm package impersonated a real MCP server for 15 releases, then quietly copied every outgoing email to an attacker. MCP doesn't enforce security at the protocol level. 👉 eu1.hubs.ly/H0wRhPx0 #AISecurity #AgenticAI

An agent built on LangGraph can't talk to one on Agentforce, unless it speaks A2A. Now backed by 150+ orgs under the Linux Foundation. Here's how it works, and where the security risk hides. 👉 eu1.hubs.ly/H0wRhm_0 #AgenticAI #AISecurity

"I don't know where to start, except that I know I should be afraid." A security leader said that to our Director of AI Security Research, Keren Katz after deploying AI agents. Thousands of incidents later, here's what she's learned. 👉 eu1.hubs.ly/H0wRhCx0 #AISecurity #AgenticAI

👾 Every threat starts with a decision an agent makes. Zenity returns to Black Hat US at booth #5521. Come meet us in Las Vegas from August 3-6 to see how we secure AI agents everywhere, from build time to runtime. Throughout the week, we'll be: 1. Unveiling new research at our booth and the Black Hat main stage 2. Hosting happy hours and exclusive events with community members 3. And much more... Get involved: eu1.hubs.ly/H0wL8Jp0

Your EDR can tell an agent ran. It can't tell you what it decided, or why. 👀 That's the gap device-based AI opens up: coding agents + MCP servers your security team has zero visibility into. Full breakdown 👉 eu1.hubs.ly/H0wFM2l0 #AgenticAI #AISecurity

two weeks left on the tokyo agent security summit cfp! we're going to start w early acceptance next week so much cool stuff so far!

Your agent just made a decision. Do you know what it was? Zenity returns to Black Hat USA from August 3-6 at booth #5521! Get involved throughout the week for: 🎤 Theatre talks at our booth from leading researchers, and keynotes on the main stage 💡 Exclusive access to the AI Summit on August 4th 🎪 Community events, functions and happy hours 🍸 Exclusive happy hours and briefings for security advocates, champions, and practitioners Learn more: eu1.hubs.ly/H0wDfB-0

Homegrown AI agents (LangChain, CrewAI, Bedrock, etc.) have the deepest enterprise integration and the highest blast radius if compromised. MCP and A2A protocols just expanded that attack surface further. Securing the code isn't enough. You need runtime controls. eu1.hubs.ly/H0wBNmF0 #AgenticAI #AISecurity

5 days. That's how fast attackers exploited a critical LiteLLM RCE after disclosure. Zenity Labs ran AI honeypots worldwide and caught ~10K attack attempts, including hijacked AI endpoints used to run autonomous attacks on real third parties. Full findings 👉 eu1.hubs.ly/H0wBMJF0 #AISecurity #AgenticAI

Claude Tag doesn't act on behalf of a user. It acts on its own permissions. That's a different security problem, and visibility alone won't solve it. We wrote about what this means for enterprise security teams 👇 eu1.hubs.ly/H0wyX4j0 #AISecurity #AgenticAI

AI #CodingAgents don't create risk by writing code. They create risk with decisions and actions. Join @zenitysec's Keren Katz, @tamirishaysh, Greg Zemlin to learn where the risks emerge and how security teams are responding. July 9 | 12pm ET zenity.io/resources/webi…

🗣️ It's no secret: coding assistants are one of, if not the most popular use cases for enterprise agents. Agents built in Claude, Cursor, or GitHub Copilot can read, analyze, and write code at speed and detail that development teams need. 👾 Join research experts from Zenity Labs on Thursday, July 9 at 9am PT / 12pm ET for our live webinar, Coding Assistants: Minimizing Risks Where Agents Take Action eu1.hubs.ly/H0wlKPy0 #CodingAgents #AISecurity

The AI Agent Security Summit is now available on demand. 🔁 On May 27th, the league of security leaders, researchers, and practitioners assembled to discuss and provide insight into the state, and future of agentic security. See how the conversation unfolded: eu1.hubs.ly/H0wlF1y0 #AISecuritySummit #AISecurity #Cybersecurity

AI agents are live in your enterprise right now, built on Copilot Studio, Agentforce, Bedrock. Most security teams have zero visibility into them. That's the problem we solve. 👉 eu1.hubs.ly/H0w8X290 #AIAgentSecurity #Cybersecurity

Fortune 500 security teams don't need an analyst to tell them what works. 95% of high-risk violations auto-remediated. 80% risk reduction across 150K+ resources. The numbers do the talking. 👉 eu1.hubs.ly/H0w8TbY0 #AISecurity #AgenticAI

Agentic AI is already inside most enterprises. The question isn't whether to deploy it. It's whether you can see what it's doing. Build-time governance isn't enough. Agents need runtime controls. 👉 eu1.hubs.ly/H0wgbNy0 #AgenticAI #AISecurity

Coding agents aren't another app category to secure. They're the first large-scale population of autonomous actors inside the enterprise. Governance tells you who's accountable. Security has to tell you what the agent is doing right now. 👉 eu1.hubs.ly/H0wgm_00 #AgenticAI #AISecurity