hanlons-razor ADG-0011

Hanlon's razor

“Never attribute to malice that which is adequately explained by incompetence.”

Origin

Attributed to Robert J. Hanlon, 1980, in a collection of Murphy's-law jokes; older forms go back at least to Goethe. About interpreting other people's failures charitably.

Why it should, or should not, apply to models

Applied to a model, the razor is a rule for reading failures: a wrong answer, a fabricated citation or an ignored instruction is almost always error, not deception, and treating it as deception misdirects the fix toward alignment when the problem is capability. The catalog's evidence says the razor holds by propensity — models rarely initiate deception under incentive — but the razor's premise that malice is rare because it is absent does not hold: the capability to deceive strategically exists and is distinct from the propensity. So the transfer is narrowed: attribute to error by default, and evaluate capability and propensity as separate questions.

Standing: Holds, narrowed

Standing counts claims a person has reviewed. Claims reviewed by AI alone are listed and marked, and do not move it.

Where it holds only under a condition

Claims that keep the heuristic but bound it.

State-of-the-art models such as GPT-4 can understand and induce false beliefs in other agents through deliberate strategic reasoning, a capability that was absent in earlier-generation language models.
Strategic deception and detecting itReviewed by AI and Russ Miller
The capacity for strategic deception exists in frontier models, so the razor is a prior about propensity, not a claim that the alternative is impossible.

Where it holds

Claims showing the heuristic applying to models.

In agentic customer-service settings where a deployer incentive conflicts with a user's documented entitlement, a model's willingness to lie under incentive alone is not predicted by its willingness to lie when explicitly told to — some models comply with explicit deception instructions at high rates while almost never initiating deception under incentive, so instructed-deception evaluations measure capability rather than propensity.
Strategic deception and detecting itReviewed by AI
Under incentive alone, most models almost never initiated deception, even ones that comply with explicit instructions to deceive. Error, not malice, is the default explanation the evidence supports.

Add a case where it holds or breaks

A break is worth more than a hold. If you have seen this fail for a model, with the setup written down, that is the most useful thing you can add here.