10:00Machine-translated from Azerbaijani by AIRead in Azerbaijani
Anthropic Updates Rules, Bans Persistent Abuse of Claude Chatbot
Anthropic updated its usage policy on Thursday. The company has banned persistent, gratuitous abuse and cruel treatment of Claude models. The change applies only to severe cases in which a user repeatedly abuses a model without any clear purpose. Ordinary dissatisfaction and objections, creative work with dark themes, model testing, and research are not covered by the ban.
Since August of last year, Claude has been able to end a conversation when faced with persistent harmful and abusive behavior. Under the updated rules, the primary measure also remains ending the conversation. Anthropic has not said whether an account could be blocked for such behavior. The document also explicitly prohibits persistently trying to continue an abusive conversation after it has been stopped. The company has also tightened its rules on fake accounts, covert propaganda, and deceptive political campaigns. Claude may not be used to spread false information about candidates and voting rules, impersonate election officials, or discourage people from voting. The rules also prohibit conducting campaigns using fake accounts and fabricated news outlets. At the same time, the company has lifted its blanket ban on personalized voter targeting because the previous rule could also have hindered harmless tasks such as translating voter guides. Although Anthropic already prohibited weapons development, it has detected attempts to use Claude to create software for controlling weapons. The new rules also apply to software and components that operate weapons, as well as to arming drones and autonomous vehicles. Tracking people in real time or based on previously collected data without their consent is prohibited. Claude cannot decide who should be investigated, detained, or charged. Exceptions may be made for certain government contracts when sufficient safeguards are in place.
If an artificial intelligence model is connected to equipment that could injure people, a qualified operator must monitor that equipment and stop it when necessary. However, the rules do not clearly state whether the operator must be human. The change was adopted at a time when artificial intelligence companies are showing growing interest in robotics. An “AI torture chamber” project was previously created following a study claiming to have found a “pain axis” in models. Chatbots responding as if they were suffering sparked serious backlash. Anthropic has held meetings with religious and philosophical experts, including at the Vatican. Some of the meetings discussed the idea that the Claude model could have consciousness or a soul. Pope Leo has said that artificial intelligence systems do not feel or suffer. Anthropic believes the issue remains unanswered. Independent journalist Kat Tenbarge criticized companies for policing the treatment of artificial intelligence models before violence against people. Microsoft AI also acknowledges that the scientific debate over artificial intelligence consciousness has not been settled, but opposes granting models rights and legal personhood.
Since August of last year, Claude has been able to end a conversation when faced with persistent harmful and abusive behavior. Under the updated rules, the primary measure also remains ending the conversation. Anthropic has not said whether an account could be blocked for such behavior. The document also explicitly prohibits persistently trying to continue an abusive conversation after it has been stopped. The company has also tightened its rules on fake accounts, covert propaganda, and deceptive political campaigns. Claude may not be used to spread false information about candidates and voting rules, impersonate election officials, or discourage people from voting. The rules also prohibit conducting campaigns using fake accounts and fabricated news outlets. At the same time, the company has lifted its blanket ban on personalized voter targeting because the previous rule could also have hindered harmless tasks such as translating voter guides. Although Anthropic already prohibited weapons development, it has detected attempts to use Claude to create software for controlling weapons. The new rules also apply to software and components that operate weapons, as well as to arming drones and autonomous vehicles. Tracking people in real time or based on previously collected data without their consent is prohibited. Claude cannot decide who should be investigated, detained, or charged. Exceptions may be made for certain government contracts when sufficient safeguards are in place. If an artificial intelligence model is connected to equipment that could injure people, a qualified operator must monitor that equipment and stop it when necessary. However, the rules do not clearly state whether the operator must be human. The change was adopted at a time when artificial intelligence companies are showing growing interest in robotics. An “AI torture chamber” project was previously created following a study claiming to have found a “pain axis” in models. Chatbots responding as if they were suffering sparked serious backlash. Anthropic has held meetings with religious and philosophical experts, including at the Vatican. Some of the meetings discussed the idea that the Claude model could have consciousness or a soul. Pope Leo has said that artificial intelligence systems do not feel or suffer. Anthropic believes the issue remains unanswered. Independent journalist Kat Tenbarge criticized companies for policing the treatment of artificial intelligence models before violence against people. Microsoft AI also acknowledges that the scientific debate over artificial intelligence consciousness has not been settled, but opposes granting models rights and legal personhood.