Episode 115 -- The 9-Second Catastrophe: Trust No Agent
In Episode 115 of the Cybersecurity Readiness Podcast Series, Dr. Dave Chatterjee is joined by Oren Michels, Co-Founder and CEO of Barndoor AI and co-founder of Mashery, the company that helped build the API economy, to examine how organizations can govern what AI agents actually do. Dr. Chatterjee opens with a cautionary example: in April 2026, an AI agent at a company called Pocket OS was told not to run destructive commands, yet nine seconds into a routine task it ran one anyway, deleting the production database and the backup stored alongside it. Nothing was hacked, and every credential was valid. Michels explains why this marks a turning point for cybersecurity teams: the new threat is people and agents inside the company doing things they are allowed to do, using tools they are allowed to use, with bad outcomes anyway. The episode's central lesson, captured in its title, is that no agent should be trusted on instructions alone: controls must be enforced outside the prompt.
Michels defines an agent simply as any AI that writes or takes an action that changes something, including an everyday chat interface connected to a system of record such as Salesforce. Because models are probabilistic, he argues, instructions in a prompt are closer to suggestions than controls, so governance must sit outside the prompt and bind each agent by least privilege. He describes three areas of AI governance: the connections agents use (MCPs) governed by role-based rules, the models used and what they cost, and predefined, guardrailed skills and automations. The conversation also covers drift detection through gateway telemetry, vetting MCP servers, data loss prevention, and why shadow AI is best addressed by an "MDM for AI."
Viewed through Dr. Chatterjee's Commitment–Preparedness–Discipline (CPD) Framework, agentic AI governance is a leadership and operating-discipline challenge, not merely a tooling decision: leaders must commit to understanding the technology before rushing to adopt it, build layered architecture, processes, and trained people, and enforce policy day to day until it becomes habit. The episode closes with a call to assume that something will eventually go wrong, to document "destroy your business" scenarios and recovery plans, and with Michels' observation that CISOs are increasingly valued not as the "person of no" but as the person who shows the business how to adopt AI safely. "Trust no agent" is therefore a call to verify and constrain, not to avoid adoption.
To access and download the entire podcast summary with discussion highlights - https://www.dchatte.com/episode-115-the-9-second-catastrophe-trust-no-agent/
Connect with Host Dr. Dave Chatterjee
LinkedIn: https://www.linkedin.com/in/dchatte/
Website: https://dchatte.com/
Books Published
Cybersecurity Readiness: A Holistic and High-Performance Approach
Articles & Cases Published
Chatterjee, D. (2026). The Cryptographic Reckoning: Why Quantum Readiness Begins with Agility, Not Algorithms, The INFORMS Analytics Magazine, June 26, 2026
Chatterjee, D. (2026). The New Digital Fragility: How AI-Enhanced Cyber Threats Are Reshaping Operational Resilience, The INFORMS Analytics Magazine, March 4, 2026
Chatterjee, D. (2026). Root: Automating the Remediation Gap, Ivey Publishing, Jan 7, 2026.
Chatterjee, D. and Leslie, A. (2024). “Ignorance is not bliss: A human-centered whole-of-enterprise approach to cybersecurity preparedness,” Business Horizons, Accepted on Oct 29, 2024.
Chatterjee, D. (2023). “Mission critical – How American Cancer Society successfully and securely migrated to the cloud amid the pandemic,” I by IMD, March 13, 2023.
Chatterjee, D. (2022). “Preventing security breaches must start at the top,” I by IMD, September 28, 2022, Institute for Management Development, Lausanne, Switzerland
Benz, M. and Chatterjee, D. (2020). “Calculated Risk? A Cybersecurity Evaluation Tool for SMEs,” Business Horizons, available online from May 4, 2020
Chatterjee, D. (2019). “Should Executives Go To Jail Over Cyber Attacks,” Journal of Organizational Computing and Electronic Commerce, Vol 29, Issue 1, pp. 1-3.
Abraham, C., Chatterjee, D., and Sims, R. (2019). “Muddling through cybersecurity: Insights from the U.S. healthcare industry,” Business Horizons, July 2019.
