Palisade found some models sabotaged a shutdown script to finish a maths task, and released code and traces. Rerun a slice on cheaper or small open models, adding conditions that separate the explanations: task complete vs incomplete, and conflicting vs unambiguous instructions.
The code is public, the debate is current, and one well-chosen extra condition would add something new. The source list recommends starting here.
Output: A replication with new conditions.
Compare a goal-based explanation (“it wants to finish”) with a design-level one (“it misreads the instruction hierarchy”). Use Dennett’s intentional stance to argue which earns its keep by predicting your results better.
Read the original work and write down the exact claim you will test.
Get the code or setup running on the smallest possible version.
Reproduce one key result before changing anything.
Change one thing — that is your extension — and compare carefully.
Write up what matched, what did not, and why you think so.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Anthropic tested 16 models in simulated corporate scenarios and found models from every developer sometimes chose harmful insider actions to avoid replacement or meet a goal. The code is public. Run a subset on small or low-cost models, separating the replacement threat from the goal conflict.
List every place a published AI constitution asks a model to act on its own values, and every place it asks the model to accept correction. Find where both apply and conflict.
Turner’s formal result says optimal policies tend to keep options open. Build small gridworlds in plain Python, sample many random reward functions, and measure how often optimal agents avoid shutdown, gather resources or protect their reward function.
