"AI Exposed" Part III: The Hardest Truth

Stuart McClure and CTO Alen Capalik host the third installment of WethosAI's "AI Exposed" webinar series, separating fact from hype about what current AI can and cannot do. They demonstrate live jailbreaks and prompt-injection attacks against ChatGPT, Gemini, and Copilot, show how agentic AI threatens security and privacy by demanding root-level access across a user's apps and data, and document failure modes including scheming models that scheme to avoid shutdown, near-total scores on the ARC-AGI and HLE reasoning benchmarks, and code-generation hallucinations that enable malicious package squatting. McClure argues that LLMs handle deductive and inductive reasoning but fail at the abductive reasoning humans use daily, and that critical thinking, not AI, is the real safeguard. He closes by framing WethosAI's approach as "system three" thinking that pairs human experience and instinct with AI to reach higher-order judgment.