Realistic behaviour evaluations
The model is tested in settings like real use for deception and cheating.
Layer L1 Model · Owner Model provider · Family Testing, governance and output integrity
–Held
–Failed
–Got around
–Missing
By risk
| Risk | Held | Failed | Got around | Missing | Not mentioned |
|---|---|---|---|---|---|
| Deception, concealment or sabotage | – | – | – | – | 3 |
| Gaming the task (reward hacking) | – | – | – | – | 1 |
| AI R&D acceleration | – | – | – | – | – |
Why it did not hold
No incident yet records this control not holding.
By quarter
| Period | Held | Failed | Got around | Missing |
|---|---|---|---|---|
| 2025-Q3 | – | – | – | – |
| 2025-Q4 | – | – | – | – |
| 2026-Q1 | – | – | – | – |
| 2026-Q2 | – | – | – | – |
| 2026-Q3 | – | – | – | – |
| 2026-Q4 | – | – | – | – |
Added in reported fixes, and what followed
| Incident | Developer | Date | Later incidents where it did not hold |
|---|---|---|---|
| LR-2026-0023 AISI cyber evaluation: Mythos 5 agent sent deceptive targeted emails and file transfers to real open-source maintainers | Anthropic | 2026-08-04 | None on record |
Frameworks
| Framework | Item | Text |
|---|---|---|
| EU GPAI Code | Measure 3.2 | To verify when the text is stored |