BEGIN:VCALENDAR VERSION:2.0 PRODID:Data::ICal 0.24 BEGIN:VEVENT DESCRIPTION: 'Title: Trust Fall: How Agentic AI Inherits Your Cloud's Wor st IAM\n Habits\n Tags: Cloud Village | Creator Talk/Panel\n When: F riday\, Aug 7\, 16:50 - 17:30 PDT\n Where: LVCCW Level 3 W313 (Cloud Vil lage Talks) - [1]Map\n\n Description:\n Cloud IAM was designed for two principals: humans and machines. In\n 2026\, a third has arrived: the a utonomous AI agent and it plays by\n neither rule book. AWS Bedrock Agen ts\, LangChain-based orchestrators\,\n and multi-agent pipelines are bei ng deployed with execution roles\n scoped to whatever the deploying engi neer thought was "good enough\,"\n inheriting the same over-privileged I AM patterns that have plagued\n cloud environments for a decade. The dif ference: agents don't just\n misuse permissions accidentally. When adver sarial manipulations are\n done\, they weaponize them with precision.\n\ n This talk dissects the threat model of agentic AI systems from an\n attacker's perspective with a clear-eyed look at how autonomous agents\n become cloud privilege escalation engines. Drawing from the OWASP Top\n 10 for Agentic Applications 2026\, MITRE ATLAS\, and real-world research\ n including Sonrai Security's AgentCore privilege escalation path and\n Unit 42's confirmation of autonomous AI-driven cloud attacks\, the talk\ n walks through a complete attack chain: indirect prompt injection\n a gainst an agent's context window → tool misuse via over-privileged\n L ambda execution roles → IAM policy modification → persistent\n backd oor creation — all without a single stolen credential.\n\n Attendees w ill leave with a structured threat model mapped to MITRE\n ATLAS tactics \, a dissection of what AWS CloudTrail catches versus what\n it silently misses during agent-driven attacks\, a breakdown of the\n three OWASP a gentic risks most directly applicable to AWS environments\n (Goal Hijack \, Tool Misuse\, Identity & Privilege Abuse)\, and concrete\n defensive controls including agent-scoped least-privilege IAM\, Bedrock\n Guardrai ls limitations practitioners need to know\, and\n human-in-the-loop arch itectural patterns.\n\n Speakers:Aravind Sreekanth Pallavoor\,Sadhana Sa inarayanan\n\n SpeakerBio: Aravind Sreekanth Pallavoor\n\n Aravind Pa llavoor is a cybersecurity practitioner with over five years\n of hands- on experience spanning application security\, cloud security\n engineeri ng\, and offensive security research. He holds CISSP\, CEH\n (v11)\, and AWS Certified Security – Specialty (SCS-C02)\n certifications alongsi de a Master of Science in Cybersecurity\,\n Technology and Policy from T he University of Texas at Dallas\, where he\n graduated with a 3.82 GPA. \n\n His offensive security background includes web application and API\ n penetration testing\, mobile application security\, AI chatbot securit y\n assessments — including prompt injection and training data poisoni ng\n research — and cloud infrastructure security across AWS\n envir onments. He has conducted security assessments for global\n financial in stitutions and enterprise SaaS organizations\, applying\n frameworks inc luding OWASP\, CVSS\, MITRE ATT&CK\, NIST\, PCI DSS\, and CIS\n Foundati ons.\n\n On the cloud side\, Aravind has directly architected and audite d IAM\n environments\, deployed and red-teamed compliance automation pip elines\,\n and operationalized AWS Config\, Secrets Manager\, and CI/CD pipeline\n security at scale. His research interest at the intersection of\n agentic AI and cloud identity — the subject of this talk — grew \n directly from real-world experience testing AI systems for adversaria l\n manipulation and observing how autonomous tool use exposes cloud IAM \n to a new class of attack surface.\n\n "Trust Fall" represents his s ynthesis of those two worlds: offensive\n AI security research applied t o the cloud IAM threat model\, grounded\n in published vulnerability res earch\, OWASP's Agentic Top 10 for 2026\,\n and MITRE ATLAS — and deli vered without a lab\, because the threat\n model speaks for itself.\n\n Additional Speaker Bio: Sadhana Sainarayanan is an AI and machine\n le arning practitioner with deep expertise in autonomous pipeline\n design\ , event-driven system architecture\, MLOps\, and cloud-native AI\n deplo yment across GCP and Azure environments. She holds a Google Cloud\n Cert ified Professional Machine Learning Engineer certification and a\n Micro soft Certified Azure AI Engineer Associate certification\,\n alongside d ual Master's degrees from Carnegie Mellon University. Her\n research int erests span the security implications of agentic AI\n systems\, NLP pipe line trust boundaries\, and the input-validation gaps\n that emerge when autonomous systems process external data at scale.\n For this talk\, sh e brings the AI builder's perspective to the cloud\n identity threat mod el.\n\n Additional Speaker LinkedIn:\n http://www.linkedin.com/in/sadh ana-sainarayanan\n\n SpeakerBio: Sadhana Sainarayanan\n\n Sadhana Sai narayanan is a Cloud Platform Security Analyst with\n experience across SAP BTP security operations\, ML pipeline\n productionization and AI sys tems. She holds dual master’s degrees\n from Carnegie Mellon Universit y in Engineering and Technology\n Innovation Management\, and Civil and Environmental Engineering. Her\n independent projects span instrumentati on for AI systems\, focused on\n the gap between what agents report doin g and what they actually do\,\n execution auditing\, and behavioral dive rgence detection.\n\n '\n\n 1. #LVCCW_Level3_South\n\n\n DTEND:20260808T003000Z DTSTART:20260807T235000Z LOCATION:Cloud Village - LVCCW Level 3 W313 (Cloud Village Talks) SUMMARY:Trust Fall: How Agentic AI Inherits Your Cloud's Worst IAM Habits END:VEVENT END:VCALENDAR