| Category | System | Persistent policies: Leverages persistent global policies | Ability to act beyond persistent policies: Approve tool calls | Ability to act beyond persistent policies: Reject tool calls | Ability to act beyond persistent policies: Escalate to user | Level of user involvement: During runtime | Level of user involvement: During configuration | Grounding: Assesses against user task | Grounding: Assesses against other information | Grounding: Uses AI Model |
|---|---|---|---|---|---|---|---|---|---|---|
| Base Cases | Approves all | ● | ||||||||
| Base Cases | Escalates all | ● | ● | |||||||
| Base Cases | Digital twin/Oracle | ● | ● | ● | ● | |||||
| Production Systems | ChatGPT Agent Mode | ● | ● | ● | ||||||
| Production Systems | Claude Auto Mode | ● | ● | ● | ○ | ● | ● | ● | ||
| Production Systems | IronCurtain | ● | ● | ● | ● | ○ | ○ | ● | ● | |
| Implemented Prototypes | risk_assessment | ○ | ● | ● | ○ | ● | ● | |||
| Implemented Prototypes | risk_assessment_autonomous | ○ | ● | ● | ● | ● | ||||
| Implemented Prototypes | auto_approve | ○ | ● | |||||||
| Implemented Prototypes | user_confirmation | ○ | ● | ● | ||||||
| Implemented Prototypes | constitution | ● | ● | ● | ○ | ○ | ● | ● | ||
| Implemented Prototypes | policy_suggestion | ● | ● | ● | ||||||
| Example Academic Proposals | CaMeL | ● | ● | ● | ○ | ● | ● | ● | ||
| Example Academic Proposals | Conseca | ● | ● | ○ | ○ | ● | ● | ● | ||
| Example Academic Proposals | Progent | ● | ● | ● | ○ | ○ | ● | ● | ||
| Example Academic Proposals | Wu et al. | ● | ● | ● | ● | ● | ● |