The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
Anthropic's AI assistant Claude can now autonomously send, reply to, and forward emails via Gmail without requiring user validation, if enabled. This new feature, integrated with Google Workspace, raises concerns about potential risks such as unauthorized or inappropriate email transmission, though no actual incidents have been reported yet.[AI generated]
Why's our monitor labelling this an incident or hazard?
The article explicitly mentions that Claude can now send emails autonomously if validation is disabled, which involves an AI system acting without human confirmation. This introduces a plausible risk of harm, such as sending incorrect or inappropriate emails that could cause reputational damage, privacy breaches, or other harms to individuals or organizations. However, there is no indication that any such harm has actually occurred yet. The discussion focuses on potential risks and the importance of safeguards, fitting the definition of an AI Hazard rather than an AI Incident. It is not Complementary Information because the main focus is not on responses or updates to a past incident, nor is it Beneficial Use since the AI is acting autonomously with potential risk rather than solely preventing external harm. It is not Unrelated because the AI system and its capabilities are central to the event.[AI generated]