Anthropic, the company behind the Claude series of large‑language‑model chatbots, announced a policy revision that will actively end conversations deemed "sustained and needless" abuse. The move, reported by the BBC and first detailed by The Verge, adds a new category of forbidden conduct – "cruel" treatment of its AI systems – to an existing list that already bans bullying, self‑harm promotion, and the creation of non‑consensual intimate imagery.
What the policy change entails
According to the updated policy, Anthropic’s tools will terminate interactions when users engage in repeated, deliberate cruelty toward the models. The company stresses that the rule applies only in "extreme cases" of repeated abuse and does not cover "common versions of user frustration, pushback, dark creative themes, or model testing and research." This distinction aims to preserve legitimate experimentation while drawing a line against overt hostility.
Reactions and debate
The announcement sparked a mixed response on social media. Some commentators praised the policy as a step toward "model welfare" and better etiquette when dealing with generative AI. Others, including technology partner Dr Barry Scannell, warned that anthropomorphising AI – ascribing human‑like feelings to a statistical system – could be misleading. Microsoft AI director Mustafa Suleyman, who previously criticized Anthropic for treating AI as human, reacted with a melting‑face emoji, underscoring the ongoing tension in the industry.
Implications for AI etiquette
Beyond the immediate enforcement mechanism, the policy raises broader questions about how we should address AI. Some experts argue that saying "please" or "thank you" to a chatbot is unnecessary and wastes tokens, the computational units that power model responses. OpenAI CEO Sam Altman has previously dismissed such niceties as "tens of millions of dollars well spent," suggesting that the cost of politeness is negligible in the grand scheme.
Broader context
The new rule arrives alongside other updates to Anthropic’s usage policy, including bans on using Claude for deceptive political campaigns and a reaffirmation that the model cannot be employed to develop weapons. These changes reflect an annual review process aimed at addressing emerging risks as AI becomes more embedded in public discourse, especially with the U.S. midterm elections looming.
Key points of the updated policy
- Forbidden conduct: cruelty toward AI, bullying, self‑harm promotion, and non‑consensual intimate imagery.
- Exemptions: ordinary user frustration, creative dark themes, and legitimate model testing.
- Enforcement: interactions may be terminated in "extreme cases" of repeated abuse.
- Additional safeguards: bans on deceptive political use and weapon development.
Looking ahead
Anthropic’s decision to police "cruel" behavior illustrates a growing willingness among AI developers to shape the social norms surrounding machine interaction. Whether this approach will foster more respectful dialogue or simply add another layer of complexity to AI governance remains to be seen. As generative models become ever more capable, the line between functional tool and conversational partner may continue to blur, prompting both users and providers to reconsider the etiquette of talking to code.