The AI Ferrari Needs Brakes
summary
Claims a prompt-based assistant fails about 1% of the time and gives concrete incidents: Air Canada forced to honor a chatbot refund, a Chevrolet dealer bot generating fake legally binding offers. Describes a deterministic Dendrite gateway that blocks API execution until human approval, circuit breakers, token budgets, and sovereign VPC deployments. Offers an open-source Apache‑2.0 repo for an intercepting gateway and a 0→4 maturity path to enforce human-in-the-loop and prevent unauthorized refunds, deletes, or runaway token costs.
AI-written summary · about
chapters (11)
- 0:00AI needs brakes like Ferrari
- 1:13Determinism vs probabilism
- 5:09AI failures and costs
- 8:43Need for gateway layer
- 12:36Autonomy is a myth
- 15:30Agentic architectures and risks
- 17:17AI governance components
- 21:13AI governance maturity stages
- 23:41Reaching enforcement control
- 27:13StackMint for AI control
- 28:37Open-source governance
related talks
18mDefending the unknown using AI and cybersecurity
30mAI Agents Unchained
54mA Practical Guide to AI Threats & Mitigation Strategies
24mBots & Humans: Maintaining AI agent and automation in an AI-bot web
24mHow to Build Trustworthy AI
01h 01mKeynote: It AI-n't What You Think!
24mBuilding Trustworthy AI Authentication & Authorization for AI Agents
14mBest Practices for Accelerating Your AI Journey