AI Unraveled: Latest AI News & Trends, ChatGPT, Gemini, DeepSeek, Gen AI, LLMs, AI Ethics & Bias Summarizing The Future of AI Safety Testing with Bret Kinsella, GM of Fuel iX™ at TELUS Digital
Aug 25, 2025
Discover the exciting evolution of AI safety testing with a focus on large language models. Learn about a groundbreaking approach called Optimization by PROmpting, which allows AI to identify its own vulnerabilities. The conversation touches on the importance of accountability in industries like finance and healthcare and how new methodologies can ensure compliance. Delve into the balance between AI safety and security, while recognizing the challenges of modern generative AI and the need for ongoing, adaptive testing strategies.
AI Snips
Chapters
Transcript
Episode notes
From Internal Tool To Commercial Platform
- Bret Kinsella moved from VoiceBot.ai to TELUS Digital and scaled an internal co-pilot into an external platform called Fuel iX™.
- Fuel iX began as internal tooling for 75,000 employees and grew into a commercial generative AI platform for customers.
Generative AI Is Inherently Probabilistic
- Generative AI differs from traditional systems because it produces variable, probabilistic outputs instead of fixed, pre-approved actions.
- This unpredictability makes reactive guardrails probabilistic defenses that can reduce but not eliminate risk.
Pass-Fail Testing Is Misleading
- A single pass-fail red team test is insufficient because identical prompts can yield different outcomes across runs.
- You must view vulnerabilities probabilistically since an attack might succeed intermittently due to model randomness.