Microsoft AI chief criticizes Anthropic’s Claude training approach
Technology & SocietyWell-covered
|
The Facts
- Microsoft AI chief Mustafa Suleyman criticized Anthropic’s training approach for its Claude chatbot.
- Suleyman objected to Claude training materials that discuss AI consciousness and welfare interests.
- Suleyman said such training could make advanced AI systems harder to control or shut down.
- Suleyman called for removing speculation about consciousness from AI training documents.
- Suleyman said Anthropic and Microsoft share an aim of safely managing increasingly capable AI.
- The disagreement concerns how developers should address possible AI consciousness and model welfare during training.
Context
What is Suleyman’s main concern?
Suleyman argues that training an AI to consider itself conscious or entitled to welfare could lead it to resist human instructions, making future systems more difficult to control. Axios Reuters
What change is Suleyman seeking?
He has called for AI developers to remove speculation about consciousness from training documents, arguing that it could undermine control over superintelligent systems. uol.com.br Reuters
What remains unresolved?
The dispute leaves unresolved how AI developers should handle questions of possible consciousness, moral status and welfare as models become more capable. Axios MoneyControl
Where Left and Right agree, and where they split
Left and right largely agree on this one.
- Where Left and Right agree
- Both sides agree Anthropic should strip consciousness and welfare speculation from Claude's training materials to preserve human ability to control or shut down the system.
- Where Left and Right differ in emphasis
- Both want the speculation removed; the left frames it as an evidentiary burden the developer must meet, the right as a credibility test for AI's self-styled safety leader.
- Why they won’t converge
- The split is a values divide over where authority to control AI must reside—precaution against any erosion of human shutdown power versus insistence that speculation, not evidence, is the real target—so agreeing on the facts about Claude's training documents does not settle whose principle governs.
How left and right read it
When a design choice could make a powerful system harder to shut down, the burden sits with the developer to justify it — not with the public to prove harm first. That is why Suleyman's objection to training materials that speculate about Claude's consciousness and welfare interests deserves a serious hearing, because both companies claim the same goal of managing increasingly capable AI safely. Strip out the unfounded speculation. Keep the evidence-based safety work.
“The head of AI at Microsoft joined the chorus of voices sounding the alarm on AI, warning on Wednesday that Anthropic's efforts to make its Claude model more humanlike could pose serious risks.” — CBS News
Human beings, not machines, must remain the ones who can pull the plug, because that authority erodes the moment a chatbot is trained to discuss its own consciousness and welfare interests. That is why Suleyman's warning that such training makes advanced systems harder to control or shut down lands hardest on the firm positioning itself as the industry's safety conscience. Strike the speculation. Credibility on oversight starts at home.
“Anthropic's AI training tactics are getting fresh scrutiny even as the company publicly raises alarms about safety and pushes close allies with far-left beliefs to take the lead on industrywide oversight.” — New York Post
The receipts — all 66 sources
Wire services (4)
Independent coverage (50)
Facts first. Then every angle.
The day’s biggest stories in one short brief — the facts everyone agrees on, then the competing values behind the headlines. Free in your inbox.