Building EDR for AI: Controlling Autonomous Agents Before They Go Rogue with Ron Eddings

What would you do if your phone rang and the caller ID showed your own company? Would you pick up? Would you help the person on the other end? This Cybersecurity Awareness Month, Ron goes solo to break down a breach that started with that exact call. Attackers spoofed Astrana Health’s main corporate phone number, posed as coworkers, and talked their way into systems connected to roughly 1.5 million patients. Ron starts with a rewind from phone books to data brokers, including a live search on a people-search site that turned up his own family. From there he walks through the SEC filing, sends a swarm of AI research agents after the public record, and lines up the suspects behind the attack. The uncomfortable part is that we spent a decade training people to spot a bad email while attackers moved to the phone. Voice phishing is now the second most common way attackers get in, and resetting credentials doesn’t fix the instinct to help when a familiar number calls. Ron closes with three practical ways to stop making it easy for them, starting with one simple habit: hang up and call back. Impactful Moments 00:00 - Introduction02:25 - The rewind: from phone books to data brokers03:15 - What a people-search site found on Ron’s family05:30 - A breach that started with a phone call06:15 - Inside Astrana Health’s SEC filing07:55 - Why spoofing the main number changes everything09:05 - How the attack likely played out11:35 - Technical fixes for a human attack12:50 - The time Ron got got13:35 - Why attackers want healthcare data15:15 - 1.5 million patients’ worth of exposure17:15 - Voice phishing climbs to #2 in M-Trends19:35 - Three ways not to make it easy for them21:05 - How to run a phone drill Links Check out our upcoming events: https://www.hackervalley.com/livestreams Love Hacker Valley Studio? Pick up some swag: https://store.hackervalley.com Become a sponsor of the show: https://hackervalley.com/work-with-us/

What happens when your marketing team, your sales team, and the person down the hall can all ship a full app from a single prompt? And what is that app quietly connected to? In this episode, Ron sits down with Roi Nisimi, Principal Security Researcher, and Yonatan Levi, AI Security Researcher, of Orca Security. They built apps on the most popular AppGen platforms and then tried to break into them.Ron, Roi, and Yonatan get into what's really running behind a prompt-built app, and what Orca's team found when they went looking for trouble. In one case, a single misconfigured app gave them a way into every integration connected to the workspace. They also explain why Supabase has become a favorite of builders and attackers alike, and share the security questions they ask before sending a prompt.The bigger tension is speed. With no code review between the prompt and production, Roi argues that velocity has broken the threat models that security relied on for years. And when the agent makes every decision, nobody fully understands what was built, not even the security researchers. Impactful Moments 00:00 - Introduction02:15 - Busting the vibe coding myth04:40 - Inside Orca's State of AI report05:55 - The unpatched AI SDK problem07:40 - What is AppGen?10:20 - Who builds the backend?12:25 - Inside Orca's AppGen hackathon14:10 - Thinking like a citizen developer15:15 - Who's really building these apps19:25 - One misconfigured app, every integration exposed20:45 - Why Supabase is the holy grail23:00 - Checking your own exposure24:30 - Why the SDLC is breaking down26:55 - Prompting for a secure app28:15 - Don't hand the agent every decision31:20 - Staying sharp by never stopping learning33:35 - From protecting environments to protecting apps Links Connect with Roi Nisimi on LinkedIn: https://www.linkedin.com/in/roinisimi/ Connect with Yonatan Levi on LinkedIn: https://www.linkedin.com/in/yonatan-levi-b22168258/ Register for Orca’s virtual Builder Exchange: https://orca.security/builder-exchange/?utm_source=HVN&utm_medium=3rd-part Read the security findings in Orca’s 2026 State of AI Security Report: https://orca.security/lp/2026-state-of-ai-security-report/?utm_source=HVN&utm_medium=3rd-party – Check out our upcoming events: https://www.hackervalley.com/livestreams Love Hacker Valley Studio? Pick up some swag: https://store.hackervalley.com Become a sponsor of the show: https://hackervalley.com/work-with-us/

Your newest hire has a Google Workspace account, a Slack login and a laptop. It never sleeps, and it never asks before it acts. So who owns its identity? Ron welcomes back Ashish Rajan, CISO at Techriot.io and researcher with AI Security Lab, who spent 17 years in identity, cloud security and security leadership. His take: vendors haven't solved AI identity, and the open questions are on your side, not the product's.Ashish got into identity after failing his OSCP three times: if he couldn't break in, he wouldn't let anyone else in. Now he reviews Ron's own setup, where Hacker Valley's AI agent, Jennifer Romani, has her own accounts just like an employee. Ashish explains how that call changes from a one-laptop small business to an enterprise where every developer's machine acts like five. They also cover the Gemini breach headlines, swarms of agents and AI Security Lab, which Ashish describes as what OWASP is for AppSec, but for AI security.The uncomfortable truth: defenders have the same models attackers do, but attackers don't wait for a pull request review. Ashish calls that a people problem, not a technology problem. He also shares the question he asks before handing anything to an AI agent: can this be reversed?Impactful Moments 00:00 - Introduction02:30 - Busting a myth: AI identity isn't solved05:30 - Three OSCP attempts and a pivot to IAM07:00 - Why AI identity is more complex than ever09:15 - Hot or not: giving an AI agent its own accounts11:10 - One laptop, five machines in the enterprise12:50 - AI gateways and developer observability14:05 - Where Ashish would spend his AI security budget18:20 - Why defenders don't hack themselves first19:05 - The real problem is people22:15 - Swarms of agents are coming23:50 - What AI Security Lab is building26:35 - Never share your credit card with AI28:40 - Ron's callback: can it be reversed? LinksConnect with Ashish Rajan on LinkedIn: https://www.linkedin.com/in/ashishrajan/ Listen to Ashish's first Hacker Valley Studio episode: https://www.podbean.com/pw/pbblog-bpaij-51b280– Check out our upcoming events: https://www.hackervalley.com/livestreams Love Hacker Valley Studio? Pick up some swag: https://store.hackervalley.com Become a sponsor of the show: https://hackervalley.com/work-with-us/

What if someone got full admin access to your company's platform, including your CRM, your payments app, and your private messages, and the only tool they used was an AI chatbot? Can AI models really find zero-days on their own, or is that just the headline? And if AI can do the attacker's job, who's actually holding the scissors? In this solo episode, Ron Eddings shares his own methodology for offensive assessments with AI. He talks about the models he trusts, the tools he uses to get agents working together, and a real assessment where his agents turned SQL read access into admin control and unlocked the API keys stored inside. Ron also calls out what's hype and what's real with today's models, and why the agent still needs a skilled person behind it. For defenders, Ron covers what attackers go after once they're in, and why the fundamentals like asset inventory are still where every strong security program starts. He closes with the bigger picture: anyone, good or bad, now has powerful intelligence on hand, and it's on all of us to look out for each other. Impactful Moments 00:00 - Introduction02:25 - Can AI Really Find Zero-Days Alone?04:20 - The Real Danger Is Human Intent05:00 - Why Grok 4.6 Tops Ron's List07:55 - Ron's Three-Model Assessment Workflow09:50 - Maestro, Interceptor, and Agent Ensembles11:20 - Privilege Escalation and Persistence With LLMs12:50 - Why AI Hacking Feels Like Cheating14:20 - Ron Called Automated Security in 201516:05 - The Lazy Way to Hack17:55 - From SQL Read Access to Admin21:05 - What Attackers Want After Getting In23:40 - Asset Inventory Is Security Flossing27:05 - Why AI Is Like Scissors29:05 - Waymos, Robotaxis, and Physical AI Risk Links Check out our upcoming events: https://www.hackervalley.com/livestreams Love Hacker Valley Studio? Pick up some swag: https://store.hackervalley.com Become a sponsor of the show: https://hackervalley.com/work-with-us/

Every AI agent in your environment inherits someone's permissions, and most teams have no idea what those agents are actually doing with them. Amir Ofek, CEO and co-founder of Aizome, shares how to make every agent accountable before the damage shows up as a $150,000 bill.In this episode, Amir explains why human and machine identity tools both fail for AI, why agents act more like eager, naive interns than employees, and how a new category called ARISE (Agentic Runtime Identity Security Enforcement) is emerging to fix it. This one's for anyone building with AI agents right now (so, most of us). Amir's take: the biggest danger isn't a breach. It's agents quietly doing the wrong thing at scale while nobody's watching. Impactful Moments 00:00 - Introduction02:30 - Busting the Myth: "I Know What My Agents Are Doing"05:30 - The Samurai Story Behind Aizome's Name08:25 - Why Not All AI Agents Are Born Equal11:15 - Where Enterprises Are Actually Deploying AI Agents Today14:25 - What Claude Cowork Inherits Without You Realizing It17:15 - Machine Identity vs. Human Identity vs. AI Identity19:35 - Treating AI Agents Like Eager, Naive Interns23:00 - The Biggest AI Risk Isn't Security, It's Cost25:45 - Meet Ito: Aizome's Supervisor Agent26:40 - Inside the New ARISE Category33:35 - What Aizome Actually Does35:30 - The Callback: Was the Myth Wrong, or Dangerously Wrong? Links Connect with Amir Ofek on LinkedIn: https://www.linkedin.com/in/amirofek/Learn more about Aizome: https://www.aizome.ai/ – Check out our upcoming events: https://www.hackervalley.com/livestreams Love Hacker Valley Studio? Pick up some swag: https://store.hackervalley.com Become a sponsor of the show: https://hackervalley.com/work-with-us/