Near-perfect IIA can be achieved on randomly initialised models that cannot solve the task, suggesting causal alignment does not require task capability

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• The effect of alignment map ϕ complexity on IIA in causal abstraction is an analogue of the probing complexity–accuracy trade-offclaim0.812
  Authors connect their finding to the prior probing literature debate
• CLMAS achieves the best IIA in the causally inaccessible (No Access) direction while matching MAS in the accessible directionfinding0.789
  Demonstrates the value of the CL auxiliary loss for recovering causal alignments when one model cannot be intervened upon.
• Future models with substantially increased capabilities will exhibit alignment faking that is more consistent, robust, and harder to detecthypothesis0.779
  Extrapolation from scale-emergence finding to future risk
• Future more capable AI systems are at risk of alignment faking, whether for benign or malicious goalshypothesis0.773
  Central forward-looking hypothesis of the paper motivating the research
• Will future AI systems naturally develop the key elements (strong conflicting preferences, situational awareness) necessary for dangerous alignment faking?question0.761
  Authors identify this as the most uncertain and important question for future work
• SOO fine-tuning could be extended to align AI representations of its own goals with human user preferences, reducing misalignment by fostering coherence between self-related and other-related preferenceshypothesis0.757
  Future work hypothesis about extending SOO to direct value alignment
• Correlative methods like RSA and CKA are insufficient for determining functional similarity between neural systems; causal methods are necessaryclaim0.753
  Central motivating claim of the paper; supported by empirical comparisons showing RSA/CKA miss Markovian differences detectable by MAS.
• Alignment-faking reasoning is almost absent (~2%) when evaluating model on benign unrelated queries with the harmful-only promptfinding0.753
  Rules out prompt-level implicit priming for alignment faking independent of query content