Episode 06: Keeping AI Safe. Today. Ready for Tomorrow

In Episode 06 of the Modernise Without Compromise series, OpenText's Glenn Jacobi and CCA Software's Margaret Dine take on the question AI adoption keeps outrunning: how do we keep AI safe? They work through why data governance is the foundation of AI safety, how zero trust has to extend to AI agents as well as people, and how encryption and tokenisation let organisations unlock data for AI without giving up control of it. Agentic AI does not follow a well-defined process - it makes the process up as it goes, which puts the data feeding it squarely in the frame. This episode traces the path from 1980s automation to unpredictable decision-making, why lineage and a trusted source are what stand between you and hallucination, why sensitive data should be encrypted by default and extended to third and fourth parties, and what good AI governance looks like in financial services and government: protection by design, continuous monitoring, and a data operation centre sitting alongside the SOC. ━━━━━━━━━━━━━━━━━━━━━━━ WHAT WE COVER ━━━━━━━━━━━━━━━━━━━━━━━ 0:00 Introduction and welcome 0:25 AI is everywhere, and the automation is outrunning us 1:24 Forty years of AI: from automation to agentic 2:31 Agentic AI makes up its own process 3:21 Data security and data governance: why both, and why now 4:08 Analytics and AI: who controls the input 5:18 Hallucinations start with a source you cannot trust 6:04 What changed: compute, price and the move to cloud 7:04 Shadow IT and the data you have lost track of 7:36 Zero trust: assume nothing, verify everything, including agents 10:39 Encryption and tokenisation for AI systems 11:31 When data leaves your frontier: third and fourth parties 12:57 Encrypted by default, decrypted by exception 13:27 Privacy frameworks and AI compliance 14:57 Discovery and classification: the encryption blueprint 17:48 Good AI governance in regulated industries 18:56 The data operation centre 20:43 The quantum era: future-proofing AI 22:45 The fundamentals: governance, identity and tokenisation 25:15 Treat AI architecture and security as one entity 26:21 If you cannot trust the data, it is already too late ━━━━━━━━━━━━━━━━━━━━━━━ KEY TAKEAWAYS ━━━━━━━━━━━━━━━━━━━━━━━ AI safety is a data problem before it is a model problem - you cannot trust the output if you cannot trust the source Agentic AI makes up its own process rather than following a defined one, so governance has to be designed in, not added after launch Zero trust applies to AI agents as well as people - every agent needs a validated identity and scoped access to the data it receives Sensitive data should be encrypted by default, and that protection has to extend to third and fourth parties Encryption and tokenisation let AI use the value of data without exposing its content, and the controls stay with the key ━━━━━━━━━━━━━━━━━━━━━━━ TALK TO A CONSULTANT ━━━━━━━━━━━━━━━━━━━━━━━ Ready to make your data safe to use for AI rather than off-limits to it? Book a briefing: https://www.ccasoftware.com/contact/ Watch the full series: https://www.ccasoftware.com/fireside/ Subscribe for new episodes as they're released. #AISafety #KeepingAISafe #AgenticAI #DataGovernance #ZeroTrust #AIDataProtection #AISecurity #GenerativeAI #DataLineage #ProtectionByDesign #OpenText #CCASoftware #ResponsibleAI