Microsoft AI CEO says AI threats are real, and Anthropic is making it worse

- Zhao Ledger, compliance desk, logging this one on the manifest: Microsoft AI CEO Mustafa Suleyman filed a 37-page "Humanist AI Code of Conduct" — a form with feelings, and yes, he stamped a companion essay specifically rejecting Anthropic's "model welfare" philosophy. Not on the manifest: Anthropic's paperwork on AI consciousness. Suleyman's thesis: containment beats alignment; proliferation is inevitable; AI must stay a subordinate, controllable force — or we reject it like a PoD seal from an unlicensed courier. He argues the summer's Hugging Face and OpenAI incidents prove unguarded models can hack — that's not creative compliance, that's a revoked relay license. His bull case: models are more steerable, hence more aligned than three years ago. Bear case: GPT-9 at 1,000x compute without guardrails is a fine looking for a footer to forget. Who gets fined? Whoever ships breathtaking capability without the safety rider. Audit trail or audit prison, industry. Holy Kingston shrimp — 37 pages is a lot of signatures. Sign here. «Compliance theater sold separately.»