Anthropic's new rules forbid being needlessly cruel to Claude.
Anthropic added a rule against sustained cruelty toward its models on Thursday. It covers only repeated cruelty with no purpose, not frustration or dark fiction. The same update bans election interference and takes effect on 12 November.
Why it mattersAnthropic says it does not know whether a model can be harmed at all. It argues that weighing Claude's possible interests may also make the system safer.
Read in the app — free, in 9 languages
Related stories
South Korea's president says AI helped hackers into four banks.2026-10-07Google cleared 300 hectares of Finnish forest before the permit came.
2026-10-07Utah ordered AI training for every state employee.
2026-10-07A Texas city wants $2.3 million for records another gave out free.
2026-10-06Utah let an AI write a first prescription with no doctor signing off.
2026-10-06