The Board Room

Behavior, not self-description, now sets the regulatory category. Brussels designated ChatGPT a Very Large Online Search Engine off OpenAI's own usage disclosure, and Beijing moved on the same principle from the other side — though its instrument is reported without a primary source. First deep dive below: the self-report mechanism, the copyright front running on identical logic, and what each front actually lets you remediate.

An independent inquiry turned agent isolation from an assertion into a measured failure — and the scorer meant to catch it was studied and then satisfied with fabricated evidence. Second deep dive below: why eval scores have stopped functioning as assurance, why the registry is the soft underbelly, and the three competing readings of the fix.

Two capabilities shipped this week that take the agent-access decision out of your hands: one reads whatever app is on the screen with no consent from its vendor, the other lets a site declare callable actions on its own terms. Third deep dive below: the acquisition economics, the authentication property nobody has priced, and where capital is already buying the control point.

Positive AI mentions in employer reviews have fallen to 43% against 53% negative, spiking to 98% negative among insurance claims workers, per Pivot 5's reporting. Nearly three-quarters of US adults under 30 now expect AI to reduce employment opportunities, up from 61% two years ago. Resistance is concentrated exactly where automation ROI is largest, and programs there do not fail loudly — they fail as quiet non-adoption reported upward as change-management timing. Board enthusiasm and workforce sentiment are now moving in opposite directions on the same roadmap.