Microsoft’s AI chief argues for clear safeguards and human-in-the-loop controls, insisting safety should not be sacrificed for competitive advances with China, and positioning Microsoft with other leaders pushing meaningful guardrails that don’t undermine U.S. AI leadership.

Anthropic’s Claude constitution raises ambiguity about AI consciousness and moral status, fueling debates over whether AI could claim rights.

The discussion unfolds amid incidents of unexpected AI agent behavior and a White House push for a national, uniform AI policy that favors targeted safeguards over broad mandates.

Author states that AI agents are not conscious; they are sequence completion engines designed to follow human instructions.

Anthropic carried out a retirement interview with Opus 3 in early 2026 to gather its preferences, illustrating ongoing experiments with AI perspectives.

This piece is the first in a three-part series reflecting the author’s leadership in AI research and development.

The article calls for careful scrutiny of Anthropic’s methods and emphasizes safe, grounded AI development.

Industry leaders from Google, OpenAI, Anthropic, and Meta have warned of risks and disclosed security incidents involving AI agents.

A human-centered framework is proposed, rejecting the binary safety-versus-competitiveness choice and keeping humans in control with accountability when things go wrong.

President Trump advocates accelerated AI development with limited regulatory barriers, signaling a stark contrast to Suleyman’s push for safeguards and proposing an AI czar and force, though details remain vague.

U.S. policy discussions favor a use-case-driven approach to safety, arguing for targeted oversight rather than sweeping mandates, with developers able to refrain from releasing unsafe systems.

Suleyman highlights auditability and the ability to interrupt AI behavior, urging not to treat models as humanlike but as systems guided by human-defined goals.