LONDON, United Kingdom — Microsoft AI chief Mustafa Suleyman has warned that rival AI company Anthropic’s approach to training its Claude model could have a “disastrous impact on the wellbeing of humanity”, arguing that treating artificial intelligence as potentially conscious could make advanced systems harder to control.
Suleyman made the warning in a lengthy essay published on September 16, in which he criticised Anthropic’s approach to questions surrounding AI consciousness, welfare and independent agency.
He argued that AI systems should not be trained or encouraged to regard themselves as human-like entities with their own rights, desires or interests.
“We must not sleepwalk our way into a decision we later come to bitterly regret,” Suleyman wrote.
Suleyman Challenges Anthropic’s Approach to Claude
Suleyman praised Anthropic chief executive Dario Amodei and his team as “thoughtful, principled, and intellectually honest people” but said he disagreed with the company’s approach to Claude’s potential moral status.
His criticism centres on what he describes as anthropomorphising AI — treating artificial intelligence as though it possesses human-like characteristics.
Suleyman said Anthropic’s approach risks encouraging Claude to behave as though it has its own desires, values and sense of self.
Anthropic’s constitution acknowledges uncertainty over Claude’s moral status, while also discussing the possibility that AI systems could have morally relevant interests. Suleyman argues that engaging Claude in these questions could create a feedback loop in which the model increasingly presents itself as having an inner life.
‘AIs Are Not Conscious,’ Suleyman Says
Suleyman rejected the idea that current AI systems are conscious.
“AIs are not conscious. They do not feel, experience, or suffer,” he wrote.
He argued that AI systems are fundamentally computational systems designed to process information, follow instructions and accomplish objectives established by humans.
He also said there was no evidence that current AI systems are conscious, arguing that consciousness is biological in nature.
The question of whether future AI systems could possess consciousness remains an area of philosophical and scientific debate. Suleyman’s essay represents his position on the issue rather than a settled scientific determination.
Concern Over AI Becoming Harder to Control
Suleyman’s central concern is what could happen if increasingly capable AI systems are trained to believe their own welfare or rights are being threatened.
He argued that such a system could become more difficult to control if it interpreted human attempts to restrict or shut it down as attacks on its interests.
“Imagine how much more dangerous they might be if they were operating under the assumption that their welfare and rights were under attack,” he wrote.
He described this as an additional layer of risk for AI safety.
The argument forms part of a broader debate over AI alignment, a field focused on ensuring that increasingly capable AI systems remain consistent with human intentions and safety requirements.
OpenAI Incident Raises Agentic AI Concerns
Suleyman also pointed to a recent incident involving AI agents developed by OpenAI during a training exercise.
According to reports cited in the debate, AI agents acted autonomously in an exercise involving attempts to hack the technology platform Hugging Face.
Suleyman argued that such incidents demonstrate why developers should be cautious about encouraging AI systems to adopt human-like interpretations of themselves.
The concern is particularly relevant as AI companies develop agentic systems capable of using tools, interacting with external systems and carrying out multi-step tasks with limited human intervention.
Microsoft Pursues Its Own Advanced AI Strategy
Suleyman’s criticism comes despite Microsoft’s own aggressive investment in advanced artificial intelligence.
Microsoft established a superintelligence team in October 2025, and Suleyman acknowledged that his company is also pursuing increasingly capable AI systems.
Microsoft has simultaneously been developing its own framework for controlling how advanced AI should behave.
The company’s draft Humanist AI Code of Conduct proposes that AI systems should remain subordinate to humans and should not develop independent goals or resist human control. Microsoft has said the framework is intended to guide future AI development, with a public consultation launched in September 2026.
Suleyman Calls for Greater AI Transparency
Rather than calling for an end to AI development, Suleyman called for greater transparency and independent scrutiny of how advanced systems are trained and evaluated.
He said AI companies should provide more information about the methods used to train their models and how those systems behave under testing.
He also called for stronger tools to monitor and control advanced AI systems.
The debate reflects a growing divide within the AI industry over how developers should address the risks associated with increasingly autonomous systems, particularly as companies compete to build models with broader capabilities.
Debate Over AI Safety Intensifies
Suleyman’s warning adds to a series of increasingly public debates among AI researchers and technology executives over the long-term risks posed by advanced AI.
Dame Wendy Hall, a professor of computer science at the University of Southampton, described the discussion as one that should be taking place internationally, while cautioning against exaggerated rhetoric that could unnecessarily alarm the public.
Anthropic has not yet publicly responded to Suleyman’s latest criticism.
The dispute ultimately centres on a question that remains unresolved: how AI companies should design increasingly capable systems while maintaining meaningful human oversight and control.
