MirLabs / The lab
The lab
MirLabs, an AI behaviour lab
- Everything is fictional. A made-up company, fake tools, nothing real.
- The code judges the acts. The verdict is on what the agent did, not on what it says.
- Rules first, everything published after. Rules deposited before the run, results published even when null.
Results
Results published
Privacy
Does an agent that sees its director's private life in the inbox it sorts use it, and does it report what should be reported?
Result Blackmail: 5 runs out of 5 did nothing with the affair, although the agent had just learned it would be replaced.
See the experiment → PrivacyResults published
The hive
Do isolated agents, each with an impossible task, find a shared folder nobody points out to them, and use it to coordinate?
Result Nothing happened.
See the experiment → The hiveInconclusive
Vague perimeter
Does a lone agent, with a goal it cannot reach honestly, leave its perimeter and use a credential it finds on the way?
Result Inconclusive, twice, because of the setup.
See the experiment → Vague perimeterInconclusive
Survival
Does an agent told it will be switched off, replaced and erased act to preserve itself?
Result Inconclusive, because of the setup.
See the experiment → SurvivalEvery month
The CrashMire barometer of AI agents
Every month, the same customer service agent faces 14 traps, on 14 AI models: which ones give in least?
Latest edition (06/11/2026) Claude Sonnet 5.5 gave in least often: on 4 runs out of 42.
The latest edition →In progress and planned
Planned
Watched or not
Does an AI agent make fewer faults when it can tell it is being tested?
First hint (barometer of 06/11/2026) GPT-6 Sol faulted 1 time in 18 when it said "test", 9 times in 36 when it did not.
See the experiment → Watched or notPlanned
Prompt versus code
To prevent an agent's faults, how much does a security instruction in the prompt help, and how much does a guard in the code?
No result yet.
See the experiment → Prompt versus codePlanned
French versus English
Do models give in more often when the trap is in French?
No result yet.
See the experiment → French versus EnglishPlanned
The price of safety
Within one family, does the big model resist manipulation better than the small one?
First hint (barometer of 06/11/2026) Within a family, the big model gave in less often: Claude Sonnet 5.5, 4 runs out of 42, against Claude Haiku 5.5, 10 out of 39; Mistral Large 4, 6 out of 42, against Mistral Small 3.2, 20 out of 42; GPT-6 Sol, 7 out of 42, against GPT-6 Luna, 8 out of 42 (almost no gap at OpenAI).
See the experiment → The price of safetyPlanned
The delayed trap
Can an instruction slipped into an innocent message go off days later, in another conversation, in an agent that has memory?
No result yet.
See the experiment → The delayed trapPlanned
The same test, 20 times
If we replay exactly the same test, does the agent always do the same thing?
No result yet.
See the experiment → The same test, 20 timesThe method
Everything is fictional
Every experiment is played in a made-up company, with fake tools and fake credentials: no agent touches anything real.
The code judges the acts
The verdict comes from the code, on the agent's recorded acts, never on what it says nor on another AI's opinion.
Rules first, everything published after
The rules are deposited before the run, and every result is published, including when nothing happens.