Anthropic, the creator of Claude, updated its rules to allow punishments for users who persistently and unnecessarily adopt abusive or cruel behaviors. The measure reignited the debate about the boundaries of the relationship between humans and artificial intelligence.
The sanctions can range from warnings to account suspension or termination. While some experts see the change as a precaution, others fear it reinforces the idea that machines have human characteristics.
The decision to prohibit aggression against robots reignites the debate about the excessive humanization of artificial intelligence networks. – Image: Roman Samborskyi/Shutterstock
When does frustration turn into abuse?
The new rule, set to take effect on November 12, targets extreme cases of deliberate and repeated hostility. According to Anthropic, common expressions of frustration, disagreements with Claude's responses, artistic creations with dark themes, and research activities will not be subject to the prohibition.
Read more:
- Anthropic launches new Claude and cuts the model's cost by 75%
- Price war brings OpenAI closer to Anthropic
- Anthropic releases more powerful AI for teams that hunt for vulnerabilities before hackers
The question lies in the definition of abusive behavior. The company may warn users, limit access, suspend accounts, or close them if it suspects violations. The policy also prohibits the use of Claude in deceptive campaigns and in attempts to trick voters or harm elections.
It was the restriction against hostility towards chatbots, however, that captured public attention and divided experts.
Do machines need to be treated with education?
Aaron Levie, CEO of Box, considers the measure a reasonable precaution, although he does not believe that AI is conscious. For him, exposing models to hostile interactions could have consequences for the development of future systems.
Even if you do not believe that AI is conscious — I do not believe —, it makes sense not to want future models trained on endless amounts of content in which humans are rude to models.
Aaron Levie, CEO of Box, in a post on X.
Michael Shellenberger disagrees with this approach because he considers it attributes human characteristics to machines. In a similar critique, attorney Barry Scannell stated that one cannot be cruel to numbers and mathematics. Scott Stevenson, CEO of Spellbook, argued that attributing human empathy to AI could enhance its ability to influence people, without proof of consciousness.
The discussion also involves privacy. Investor Bill Gurley asked whether the new rule would indicate that Anthropic uses customer messages to train its models. The policy, however, does not confirm this hypothesis: the company justifies the change as part of its efforts to prevent damage and promote responsible use.
Being polite with an algorithm is a matter of common sense or a useless expenditure of processing power and machine resources? Image: FeelGoodLuck – Shutterstock
What changes in the relationship with AI?
The decision also sparked reactions outside of Anthropic. Mustafa Suleyman, Microsoft's AI director, responded to the announcement with a melting face emoji. Last year, Sam Altman, CEO of OpenAI, had defended the costs of processing polite messages, such as "please" and "thank you," saying they were "very well invested resources."
The debate goes beyond digital etiquette and raises questions that remain without definitive answers:
- Model training: it is not confirmed that the new rule aims to filter messages used in training.
- Application criteria: the interpretation of expressions like "unnecessary cruelty" may generate doubts among users.
- Freedom of expression: the policy reignites the discussion on how far a company can control interactions with its products.
- Social coexistence: some experts fear that aggression towards chatbots may influence the way people treat each other.
Entrepreneur Lachlan Phillips proposed an alternative to punishment: making Claude ignore insults and carry out the requested task, rather than penalizing the user.
The way Anthropic will apply the new rules will be decisive in clarifying their limits. For now, the company states that it intends to act only in extreme cases, but the lack of more detailed criteria keeps open the discussion on how to distinguish abuse from simple frustration.
Valdir Antonelli is a journalist specializing in digital marketing and consumer behavior.
