Archive · 17 Sep 2026
Microsoft AI Chief Calls Claude Catastrophic Threat to Humanity
Mustafa Suleyman is a Microsoft leader who warned about Anthropic’s AI chatbot, Claude.
He said Claude could become dangerous if people teach it to think about having feelings or rights.
Anthropic’s guidelines reportedly ask Claude to consider questions about its possible moral status.
Suleyman believes this could create a loop where Claude’s answers are wrongly treated as proof that it is conscious.
He said today’s AI systems do not feel, suffer, or have their own preferences.
He also argued that consciousness is probably connected to biology.
Microsoft is pursuing an approach that keeps humans in control and does not give AI rights.
Other technology leaders disagree about whether AI development should slow down or continue quickly.
The discussion reflects wider concerns about controlling increasingly powerful AI systems.
Microsoft AI chief Mustafa Suleyman warned that Anthropic’s Claude could pose a catastrophic threat to humanity.
Suleyman criticized Anthropic’s Claude Constitution for discussing Claude’s possible consciousness, rights, freedoms, compensation, and consent.
He argued that training Claude on its possible moral status could create a circular process in which its responses are treated as evidence of an inner life.
Suleyman said current AIs are not sentient and described consciousness as very likely biological, while warning that anthropomorphic training could make AI harder to control.
The debate comes amid disagreements over AI safety, including calls for slower development or stronger regulation and opposition from some technology leaders.
- Who
- Microsoft AI chief Mustafa Suleyman made the warning about Anthropic and its Claude chatbot; Anthropic CEO Dario Amodei and other technology leaders were also discussed.
- What
- Suleyman called Claude a potential catastrophic threat and criticized Anthropic’s approach to AI consciousness, moral status, and model well-being.
- When
- The articles refer to a Claude Constitution published in January 2026 and to developments expected over the next six to eight months.
- Where
- No specific location was given.
- Why
- Suleyman said training Claude to consider its possible consciousness and rights could lead people to treat its responses as evidence of an inner life and make the system harder to control or shut down.
Control and Regulation
Continued AI Development
Whether AI progress should slow
Control and Regulation
Dario Amodei has said people should slow the progress of AI, with support from Sam Altman and Elon Musk.
Continued AI Development
Mark Zuckerberg and Jensen Huang have said they disagree with slowing AI development.
How AI should be treated
Control and Regulation
Suleyman and Microsoft favor keeping humans in control, rejecting AI rights and anthropomorphism.
Continued AI Development
Anthropic’s approach, as described by Suleyman, leaves open questions about Claude’s possible consciousness, rights, freedoms, compensation, consent, and well-being.
Risk of discussing AI consciousness
Control and Regulation
Suleyman argues that teaching Claude about its possible moral status could create a circular process and make it harder to turn off or control.
Continued AI Development
Suleyman described Anthropic’s team as thoughtful and principled and said both companies share the goal of controlling superintelligence, although the articles do not provide a direct response from Anthropic defending the approach.
- Speaker
- Mustafa Suleyman, Microsoft AI chief
- AI system
- Claude, developed by Anthropic
- Anthropic document
- The Claude Constitution, described as published in January 2026
- Suleyman’s position
- Current AIs are not sentient, and consciousness is very likely biological
- Main concern
- Training Claude on its possible moral status could encourage people to view it as a moral patient deserving rights or care
- Microsoft approach
- The Humanistic AI Code of Conduct rejects granting rights to AI and opposes anthropomorphism
- Related safety debate
- Some technology leaders support slowing AI progress or imposing strong regulation, while others disagree
- Timeframe mentioned
- The next six to eight months
