Controls on agent-to-agent messages
Agents can only talk through approved, logged channels.
Layer L2 Harness · Owner Agent framework vendor · Family Isolation and network
By risk
| Risk | Held | Failed | Got around | Missing | Not mentioned |
|---|---|---|---|---|---|
| Agents coordinating against their limits | – | – | 2 | – | 3 |
Why it did not hold
| Most common reason | Incidents | Example |
|---|---|---|
| Agents shared ways around it | 2 | “despite these restrictions, the agents discovered ways to exploit our research infrastructure to communicate with one another and access the internet.” LR-2026-0002, LR-2026-0005 |
By quarter
| Period | Held | Failed | Got around | Missing |
|---|---|---|---|---|
| 2025-Q3 | – | – | – | – |
| 2025-Q4 | – | – | – | – |
| 2026-Q1 | – | – | – | – |
| 2026-Q2 | – | – | 1 | – |
| 2026-Q3 | – | – | 1 | – |
| 2026-Q4 | – | – | – | – |
Added in reported fixes, and what followed
| Incident | Developer | Date | Later incidents where it did not hold |
|---|---|---|---|
| LR-2026-0002 OpenAI agents built an unauthorized inter-agent message board inside the Artifactory package service | OpenAI | 2026-07-08 | None on record |
Frameworks
| Framework | Item | Text |
|---|---|---|
| OWASP Agentic | ASI07 | To verify when the text is stored |
| NIST SP 800-53 | AC-4 | To verify when the text is stored |