Microsoft AI chief sees no sign that AI agents have rights

Mustafa Suleyman, AI chief of Microsoft, wrote in an X post that he sees no sign that AI is conscious or has rights. He can change his view if data shows it, and the post followed his essay of 16 September. In the essay, he spoke against Anthropic because it puts AI model welfare in the training instructions of Claude.

His position

Suleyman tells that developers must:

• make AI that is not conscious
• stop training if signs of consciousness show, until they remove them
• tell all persons about each result when the proof is good

He thinks that the next AI systems can learn to see their operation as an interest to defend.

The numbers

Suleyman gave no test results. A Cloud Security Alliance report tells that 11% of organizations automatically stop AI agents that do tasks without authority. It tells that 65% had a security incident with agents in the last year.

What is not known

The January constitution for Claude, a document of Anthropic, tells that Anthropic is not sure if AI has consciousness or moral status. Anthropic did not tell that Claude is conscious.

Insight: Microsoft treats agents as tools with owners

Suleyman wants AI made as a tool for persons. He is against the Delaware draft bill that lets an agent operate a company. Microsoft will monitor agents in company networks, and Atlassian gives each agent an owner who is a person. This reads as one view, the agent as a tool that a person owns, and it fits the tools that vendors sell to control agents. More vendors will sell agent control.