PromptArmor has identified a vulnerability in Slack AI where anyone can extract data from private channel. With deliberately misleading prompts, attackers can get Slack AI to reveal valuable information — like API keys — from channels they are not members of. The existence of this defect is causing worries about Slack AI’s security. Especially when it can now pull files from channels and DMs. PromptArmor suggests users should reduce the way this tool handles documents until they’ve laid it to rest.