#22 (EN): The Big Bluff

This episode dissects the so-called AI constitution signed at the White House on September 29th, officially the White House Accord on Super Intelligence. Daniel read the actual one-page document and the accompanying executive order renaming AI to Super Intelligence in US federal usage, and walks through why the agreement looks like corporate governance but carries no enforceable obligations. The discussion covers who signed, Google, Anthropic, Meta, xAI, Nvidia, and who conspicuously didn't, Microsoft, Amazon, and Salesforce, the three hyperscalers most companies actually run their AI workloads on. Daniel and Nova compare the document's four-layer control structure to the three-lines-of-defense model used in corporate audit, explain why the wording, should instead of must, no defined frontier model, no named reviewer, no binding consequences, makes it toothless, and translate all of this into three concrete action areas for IT leaders: AI strategy under the EU AI Act, AI security and shared responsibility for agent permissions and data access, and provider diversity to avoid lock-in.

The episode closes with a practical four-question bluff check IT leaders can use in any vendor conversation: is the promise contractual, who audits it and against what standard, what happens on breach, and who receives the audit report. Daniel also shares a real experience of a major AI provider changing subscription billing overnight, reinforcing why self-regulation from AI vendors should not be mistaken for a security strategy.

Key topics:

- What the White House Accord on Super Intelligence actually says versus what headlines claimed

- Why Microsoft, Amazon, and Salesforce did not sign, and what that means for companies running Azure or AWS

- The four-layer control model compared to the three-lines-of-defense audit framework used in enterprises

- Specific wording gaps: should versus must, no definition of frontier models, no named external reviewer, no enforceable consequences

- The executive order renaming AI to Super Intelligence in US government usage and why that terminology shift raises expectations

- Three concrete areas IT leaders must own regardless of the paper: AI strategy under the EU AI Act, internal AI security and agent permissions, and provider diversity

- A four-question bluff check for evaluating any AI vendor promise: commitment, review standard, consequences, transparency

- Lessons from a real billing change by a major AI provider and why contracts beat trust

Key takeaway: A voluntary industry statement with no enforceable standard, no independent reviewer, and no customer-facing reporting does not replace your own AI governance. Treat vendor security promises as unverified until they appear in a contract, get checked against a named standard, carry real consequences, and get reported to someone other than the vendor's own board.