Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Anthropic Adds Clause Prohibiting Persistent Abuse of Claude

AI summary

Anthropic added a clause in early October to its user agreement prohibiting users from persistently and unnecessarily insulting or otherwise mistreating models. The report says Claude could already terminate conversations in response to sustained hostile attacks, while the new rule could lead to permanent bans across Anthropic’s products in extreme cases. Anthropic describes the clause as a low-cost precaution and cites its own experiments suggesting that emotionally negative prompts may affect model behavior. Researchers quoted in the report say evidence on how hostile or rude wording affects performance remains mixed; some studies have found that sharp questions can sometimes produce better results.

In-depth insight

The clause adds account-level consequences for persistent abuse and links user behavior to possible model-safety risks. The reported evidence does not establish that Claude has emotions or subjective awareness.

10月初,AI巨头Anthropic在用户使用协议中新增条款。除了禁止用户利用Claude研发生物武器、售卖违禁药品之外,新增了一条规定:用户不得“持续、无必要地对模型实施辱骂、恶意对待行为”。

早在去年,Claude就具备主动终止对话的能力,面对持续恶意攻击的用户可以断开会话。而新规落地后,极端情况下用户可能会被永久封禁,无法使用Anthropic旗下全部AI产品。

这项新规一出,引来不少嘲讽。微软AI负责人穆斯塔法・苏莱曼直接用融化脸表情做出回应。不少人调侃,这家公司把AI当成拥有情绪、具备自我意识的生命体。

Anthropic预计本月晚些时候登陆美股IPO,该公司长期研究模型福祉,还专门设有团队研究模型“心理状态”。但专家表示,这条政策背后,不只是单纯不想让人们“欺负一台机器”。

计算机与AI领域一条经典底层原则:系统输出质量,受限于输入质量。老话讲,输入是垃圾,输出也只会是垃圾。

#Anthropic#Claude#Claude Sonnet 4.5

Original · 4 min readRead the original
Group chat

No comments yet. Start the conversation.