Activation includes ordinary references to notes, memory, and decisions without requiring Cobrain context.
- Category
- Trigger Abuse
- Confidence
- 98% confidence
- Finding
The activation description includes generic references to notes or memory and anything considered worth keeping, without requiring an explicit Cobrain request or established user preference. Combined with instructions to save information immediately, this can cause unrelated or sensitive conversation content to be persisted to a remote service and made available to other connected AIs without task-specific authorization. Persistent memory is the skill's legitimate purpose, which reduces suspicion of malicious intent, but does not establish consent to save arbitrary conversation content.
- Content
