ARTROPROPICA has announce new capabilities That will allow some of the latest models to stop the conversation of what corporate displayed is displayed as the “seldom, consisting of continuous extremities or interactions.” Strikingly, an anticropic said to do this to avoid protecting human users, but their own AI models.
So obviously, the company does not claim that Claude Ai Claude models are sent or can be harmed to the conversation with the user. In its own words, an anticropic remains “unsure about the potential moral status of Claude and other llms, now or at home.”
However, an announcement value for The final program is made to study what is called “model welfare“And said the essentially to take the approach only based on,” worked to identify and apply cheap interventions to reduce the risk to the well-being able to do. “
The latest change is now limited to Claude Opus 4 and 4.1. And again, it will only happen in the “extreme case,” such as “demands from the user to sexual content involving small-scale consisting of a scale or terror.”
While these types of requests may be able to cause legal or public problems for the anthropic of their own (witnessing a new report on how ChatGTs may be able to empower or contribute to users’), the company said that the pre-deployment test, Opus Opus 4 shows “strong options for” respond to the request and “stress patterns” when they do.
As this new conversation ability, the company said, “In all cases, Claude only uses the ability because many attempts have failed, or when the user is clearly asked to talk.”
AROTHIPIK also said Claude had “directed to not use this ability in the case of users who could dangerously dangerous self or others.”
TechCrunch Events
San Francisco
|
October 27, 2025
When Claude does conversation, an anticropic says the user will still start new conversations from the same account, and make a new branch of conversation.
“We treat this feature as a consistent experiment and will continue writing our approach,” said the company.