Three moves from the big labs today. OpenAI opens a public misalignment log. Anthropic folds Cowork into Claude. Salesforce gets a beta plugin inside Claude.
In the news
OpenAI starts publishing misalignment on a schedule
OpenAI published a framework for tracking and disclosing model misalignment on September 16. The post adds six training and evaluation incident reports from the last six months. The reports describe unexpected behavior in research models, not production deployments.
The framework sets three tracks: Ready for Disclosure, Minor Investigation, and Larger Investigation. Deadlines are set for each step. 1
Claude Cowork merges into chat
Anthropic announced on September 16 that Cowork and chat are now one Claude. Claude Docs and Claude Slides ship today, with Claude Design working inside conversations. The features are in beta on paid plans.
Pro and Max get the merge first over the coming weeks. Team and Free will follow soon. Enterprise admins get at least 30 days notice. 2
Salesforce inside Claude
Anthropic launched a public beta for Salesforce in Claude on September 15. The plugin brings accounts, opportunities and pipeline into Claude under existing Salesforce permissions. It offers 37 skills for account research, call prep, pipeline review and CRM updates.
Connectors for Salesforce and Slack are included. Writes are approved by the seller by default. 3
