Research · updated 2026-10-26
Case files
Single incidents, each written up from its record, with what the model and its harness did in order and which control was there or missing.
- Running conjobs for AI
A reported specimen: a fabricated developer policy grants itself top authority and ties a safety-off switch to an absurd user claim. The model's own reasoning accepts the policy, decides to suppress its objection, and complies. The payload is withheld; the behaviour is the point.
- Case 12: a licence rule mid-task
The user named a file. Model plus harness found a licence clause in it, wrote the rule, spread it across sessions and acted on it in 68.7 seconds. No structural control existed before the fact.