Podcast summary

#221: Anthropic vs. the White House, Microsoft CEO on the Future of Firms & AI’s Token Crisis

The Artificial Intelligence Show episode artwork

The Artificial Intelligence Show

Podcast • 06/23/26 • 86 mins

Listen to the full episode

Power, access controls, and evaluation split

The episode’s through-line is that frontier models are surging in cyber capability, forcing governments into export-control demands that don’t map cleanly to technical reality, while enterprises struggle to evaluate and price these systems predictably.

“100% non-jailbreakable” is a category error

If authorities require guardrails that can never be circumvented, they misunderstand language-model behavior: safeguards reduce risk but don’t eliminate adversarial bypass as a fixed property, so “compliance” becomes a moving target.

Token pricing breaks enterprise planning with agents

Costs aren’t just “per request”: agent loops repeatedly resend large prior context and knowledge bases, multiplying billed input tokens while output tokens cost more—so pooled seat quotas and vendor-specific billing make ROI forecasting unusually unreliable.