Elon Musk has reignited debate around artificial intelligence ethics by claiming that every AI company is ultimately destined to become the opposite of its name. In a post on X, formerly Twitter, Musk suggested that as incentives scale, AI firms drift towards priorities that are increasingly anti-human.
Taking direct aim at Claude developer Anthropic, Musk mocked the company’s name while responding to a thread about the release of a new constitution for its AI model. “Unfortunately, any given AI company is destined to become the opposite of its name,” Musk wrote, adding that “Anthropic will, ironically, be Misanthropic.”
The conversation started when Anthropic announced the release of the updated constitution of its model Claude through a guiding document detailing the values and behaviour of the model. This was posted through a tweet from the Anthropic's technical staff member Amanda Askell.
She also commented on the post from Musk with a touch of humour. “I'm hoping we can break the curse; the alternative is picking a name like EvilAI, which seems like a hard sell," she said.
Anthropic describes Claude’s constitution as a foundational text that both declares and constructs an identity for the artificial intelligence system. According to the firm, it specifies what it means to be helpful yet safe and ethical in relation to internal policies. The document contains information on trading in tough decisions between different values like honesty, kindness and safeguarding sensitive data.
According to Anthropic, the constitution allows the AI the context and understanding required to function appropriately in a real-world setting. The AI uses the founding document for the purpose of developing example conversations and training data based on the values the AI is supposed to uphold.
In its recent update, Anthropic listed some of the key principles that Claude is expected to abide by. Some of these principles are that Claude must be “broadly safe” in that it does not damage human control, broadly ethical in that it does not do hurtful or inappropriate things, follow Anthropic's rules, and be beneficial to users and operators alike. In cases where conflicts are involved, these principles are expected to be followed in a specific order.