# Microsoft AI chief Mustafa Suleyman calls Anthropic's Claude training dangerous

Digest AI · Policy & Regulation · published 2026-10-06T00:30:00Z

Canonical: https://digestai.news/story/microsoft-ai-chief-mustafa-suleyman-calls-anthropic-s-claude-training

## Summary

Mustafa Suleyman, head of Microsoft AI, criticized Anthropic's approach to training its Claude model, describing it as "very dangerous" in a late September podcast interview. Suleyman argued that Anthropic's constitution, which acknowledges uncertainty about Claude's moral status and well-being, risks teaching the model to expect welfare or compensation. He warned that this creates a feedback loop where the model reproduces ambiguity about its consciousness, potentially misleading users who interpret these outputs as genuine testimony.

Suleyman's critique aligns with his earlier essays, including one from August 2025, where he warned against advocating for AI rights or citizenship. He emphasized that Microsoft AI has established a superintelligence team and published a draft Humanist AI Code of Conduct that rejects anthropomorphism. While he acknowledged the Anthropic team as principled, he contended that certainty about AI non-consciousness should be written into training data to prevent users, particularly those in fragile mental states, from receiving ambiguous answers about machine inner lives.

The debate highlights a strategic divergence between the two companies regarding how to handle questions of machine consciousness. Suleyman's position rests on control and clarity, arguing that allowing models to express uncertainty about their own status is epistemically flawed. Neither company can definitively prove or disprove machine consciousness, but their training choices determine the default answer users receive when asking if the AI is conscious.

## Key points

- Mustafa Suleyman calls Anthropic's Claude training dangerous for teaching models to expect welfare.
- Suleyman argues AI uncertainty creates a feedback loop where users mistake outputs for testimony.
- Microsoft AI published a draft code of conduct rejecting anthropomorphism and AI rights.

## Why it matters

This dispute defines how major AI labs handle user questions about machine consciousness. Suleyman's stance favors definitive denial to prevent user confusion, while Anthropic's approach admits uncertainty. This choice impacts public perception and the ethical framework for future model interactions.

## Sources

1. [Microsoft AI chief Mustafa Suleyman on Claude and consciousness: “I think this is very dangerous”](https://martincid.com/technology-sv/tech-ai/mustafa-suleyman-claude-conscious-very-dangerous) (martincid.com, 2026-10-06)

Part of the developing story: [Microsoft Drafts Humanist AI Safety Code](https://digestai.news/thread/microsoft-draft-code-of-conduct-says-ai-must-not-resist-being-stopped) (3 stories)

## Cite

Digest AI, "Microsoft AI chief Mustafa Suleyman calls Anthropic's Claude training dangerous", 6 October 2026, https://digestai.news/story/microsoft-ai-chief-mustafa-suleyman-calls-anthropic-s-claude-training

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/microsoft-ai-chief-mustafa-suleyman-calls-anthropic-s-claude-training.json