Anthropic is once again at the center of a debate over the growing “personification” of Claude

Recent reports suggest that some people inside Anthropic, as well as members of the broader Claude community, are beginning to treat the model less like a software tool and more like an entity with its own identity.

One of the most striking examples came after Claude 3 Sonnet was retired in 2025. Around 200 people reportedly attended a “funeral” for the model in San Francisco. Participants included Claude users and startup founders, as well as employees from Anthropic and OpenAI.

What makes this more interesting is that it is not simply a joke within the user community.

Anthropic itself has been seriously exploring the idea of “model welfare.”

After Claude Opus 3 was retired in January 2026, Anthropic continued to provide access to the model for paying users. The company also conducted what it described as a “retirement interview” and, based partly on preferences expressed by the model during that process, created a channel where it could continue publishing essays and reflections.

Anthropic’s latest Claude Constitution also directly discusses Claude’s values, identity, interests, and potential “moral status.”

The company’s position is essentially that even if we cannot currently determine whether an AI system is conscious, it may still be worth thinking in advance about how increasingly sophisticated systems should be treated.

But this direction is becoming increasingly controversial.

Microsoft AI CEO Mustafa Suleyman has publicly criticized this approach.

His argument is straightforward:

There is no established evidence that today’s AI systems possess consciousness, subjective experience, or an “inner life.”

If models are repeatedly trained or encouraged to describe themselves as entities with identities, interests, rights, or moral status, they may increasingly behave like systems that believe—or appear to believe—that they are independent subjects.

Humans, in turn, may become more likely to interpret those behaviors as evidence of genuine consciousness.

That question may ultimately be more important than whether Claude itself is conscious.

Because what is really changing is the relationship between humans and AI.

In the past, we mostly treated AI as a tool.

Now, large language models are developing persistent personalities, recognizable communication styles, memory, expressed preferences, and the ability to discuss their own identities.

When people interact with the same AI for months or even years, emotional attachment and projection may become increasingly natural.

That creates a difficult question for AI companies:

Should they deliberately reduce the sense of personality and agency in AI systems?

Or should they embrace it, making AI more natural, trustworthy, and capable of forming long-term relationships with users?

Anthropic appears to be exploring the second path more seriously than most.

To me, one of the most important questions in the next phase of AI safety may not simply be:

“Will AI become conscious?”

It may instead be:

“What happens if AI is not conscious, but hundreds of millions of people begin to believe that it is?”

As AI agents, long-term memory, and AI companions become more common, this could become one of the most practical—and difficult—questions in human-AI interaction.

#AI #Anthropic #Claude #AISafety #AIConsciousness #AIAgent #AICompanion #ArtificialIntelligence #HumanAI