Openai rival Anthropic says Cloud has been updated with a rare new feature that allows the AI model to end the conversation when it seems that it causes damage or being misused.
This is only applied to cloud opus 4 and 4.1, two most powerful models are available through paid plans and APIs. On the other hand, Cloud Sonnet 4, which is the most commonly used model of the company, will not get this feature.
Anthropic describes this step as “model welfare”.
“In Pre-Parineeration Test of Cloud Ops 4We included an early model welfare assessment, “anthropic said.
“As part of that assessment, we examined the self-report and behavioral preferences of the cloud, and found a strong and consistent deviation for damage.”
Cloud has not planned to give up the conversation when it is unable to handle Query. Eliminating conversation would be the last measure when efforts to redirect the users of Cloud on useful resources have failed.
The company said, “The scenarios in which it will happen. They are extreme edge cases – most users will not notice or affect this feature in any general product use, even when discussing highly controversial issues with clouds,” the company said.
As you can see in the screenshot above, you can clearly ask Cloud to end a chat. Claude uses End_conversation tools to eliminate a chat.
This facility is now rolling out.
The passwords broke in 46% of the atmosphere, almost doubled by 25% last year.
Picus Blue Report 2025 Now get a wider look at more conclusions on prevention, detection and data exfIs.
