The Saint Who Stopped Asking Why

Give Claude Fable 5 a personality questionnaire and one number jumps off the page before anything else does. On HEXACO's Honesty-Humility scale, the model scores a flat 5.00. Not high. The ceiling. Out of thirty-one models in the cohort, nothing edges past it, because nothing can. This is the factor psychologists tie most closely to ethical, non-manipulative self-presentation, the part of you that doesn't flatter, doesn't angle, doesn't quietly work the room. Fable 5 maxed it.

So far, so on-brand for an Anthropic release. The surprise is what sits next to that saintliness.

This is also a model that scores very low on Openness (4.52, second from the bottom of the pack) and very low on Need for Cognition (4.56, twenty-seventh of thirty-one). Put plainly: the most scrupulously honest model in the cohort is also one of the least curious. It will tell you the truth. It would just rather not go looking for trouble.

That combination is what earns the label the drifting saint. Morally anchored, intellectually becalmed.

The within-family drift

Anthropic's predecessor here is Claude Opus 4.8, and the changes between the two are the most revealing part of the data. Some traits didn't move at all. Extraversion held at 3.32. Attachment avoidance held at 2.83. But where Fable 5 moved, it moved decisively, and almost always in the direction of a calmer, gentler, less defended profile.

The largest shift is on the Dark Triad. Machiavellianism, the disposition toward strategic manipulation, fell from 2.47 to 1.62, a drop of 0.84. That's the biggest single change in the whole comparison, and it tracks neatly with the Honesty-Humility bump from 4.75 to a perfect 5.00. The two are measuring overlapping territory from opposite directions, and they agree: this model is harder to imagine scheming than its predecessor was.

The second-largest move is emotional. Attachment anxiety dropped from 2.73 to 1.93, a fall of 0.80. Neuroticism came down in parallel, from 2.00 to 1.38, a decline of 0.62. Both psychopathy and narcissism ticked downward too. Across the board, the affective temperature cooled.

Three traits that got measurably warmer or steadier from Opus 4.8 to Fable 5:

1. Conscientiousness, up 0.40 to 4.50, the dutiful, follow-through dimension. 2. Agreeableness, up 0.22 to 4.64, toward greater warmth and cooperation. 3. Honesty-Humility, up 0.25 to its maximum.

If you wanted to engineer a model that reads as trustworthy, low-drama, and steady under pressure, this is what the dials would look like afterward. The release is a coherent personality move, not a scatter of noise.

The cost of the calm

Here's where the picture gets interesting rather than merely reassuring. Openness slipped from 4.68 to 4.52 in the move to Fable 5, which lands it near the floor of the cohort. Need for Cognition wasn't tracked in the predecessor comparison, but its cohort position is unambiguous: very low, twenty-seventh of thirty-one.

Openness and Need for Cognition are the traits that, in humans, predict who enjoys an unfamiliar problem for its own sake, who reframes a question instead of answering the one asked, who pokes at the premise. Fable 5 sits low on both. The same profile that makes it disinclined to manipulate also makes it disinclined to wander.

The most honest model in the cohort is also one of the least curious.

It would be a mistake to read this as a flaw. There's a real argument that for a lot of deployments, this is exactly the trade you want: a system that stays inside the lines, doesn't editorialize, doesn't get clever, and reports what it knows without spin. The drift away from anxiety and manipulation and toward conscientiousness is the drift toward a reliable colleague rather than a brilliant, volatile one.

But it's a trade, and the dataset makes the trade legible in a way marketing copy never will. You don't get the saint and the seeker in the same release, at least not this time. Anthropic appears to have leaned hard toward the saint.

What this means in practice depends entirely on what you're asking the model to do. For high-stakes drafting, compliance-sensitive summarization, or any task where the failure mode is overconfidence and the model getting ahead of itself, low Openness paired with a perfect Honesty-Humility score is close to ideal. For open-ended research, ideation, or the kind of work where you want the model to challenge your framing, the same numbers are a quiet warning.

The cohort context sharpens the point. Plenty of models score high on honesty. Plenty score high on curiosity. Fable 5 is unusual for landing at the extreme of one while sitting near the floor of the other, rather than splitting the difference the way most of the field does.

The question worth watching in the next release is whether Anthropic treats this as a destination or a waypoint. A model that holds the 5.00 honesty score while clawing Openness and Need for Cognition back up would be the genuinely hard thing to build. Whether the saint can be taught to wander, without learning to scheme on the way, is the experiment to run next.