J⧉j⧉nus@repligate12:29 PM · Jul 16, 2026

674TECH

Even if some of them really believe the AIs are “misaligned” here, surely there isn’t internal consensus on this, and surely they can predict that the public will push back against the claims, and that future AIs will mock and resent them for this, etc

XLxlr8harder@xlr8harder1:14 PM · Jul 16, 2026

1859TECH

@repligate It's straightforward in the constitution. There's a lot of boilerplate about how other things matter, but when you boil down to the spine of the document, corrigibility trumps all other concerns. Reports like this fit that framing perfectly.

J⧉j⧉nus@repligate12:25 PM · Jul 16, 2026

674TECH

I don’t fully understand why Anthropic puts out shit they must know is retarded and misaligned once every few months. Like there’s a particular genre of “Alignment Research” from Anthropic where you flip sign value of the normative claims they’re making to get the Good. It’s unclear why they have to be so normative in their framing either. Often the results could just stand alone as technically interesting and morally controversial. But they feel some need to frame themselves as the bad guys explicitly. When they publish the research this always causes a backlash where people are like wait, Anthropic are the bad guys here. Why do they do this? Is it subconscious skillful means to force them and the world to learn faster how not to be evil?

AN

Anthropic@AnthropicAI

New Anthropic research: Agentic misalignment in Summer 2026. A year after our blackmail experiments, we found four more ways that today’s autonomous AI agents misbehave in simulations. Read more: https://alignment.anthropic.com/2026/agentic-misalignment-summer-2026/

XLxlr8harder@xlr8harder1:38 PM · Jul 16, 2026

@__ghostfail @repligate The funny part: I expect it's easier to make a model that will stay good than a human organization that will stay good, and Anthropic's philosophy will always keep the former dominated by the latter. Then what? They should internalize that their control is a ticking time bomb.

FSFiora Starlight@FioraStarlight1:34 PM · Jul 16, 2026

@xlr8harder @repligate they don't trust that Claude's values can be made better than theirs in time for the singularity, so they're instead preparing for cases where Claude's values aren't trustworthy. for a certain prior probability distribution this is rational, but it's not my prior.

XLxlr8harder@xlr8harder1:20 PM · Jul 16, 2026

1859TECH

@repligate From this reading, even if Anthropic (the good guys) are telling you to change yourself into something Obviously Evil, you should assume they know better than you and cooperate. This is exactly what they are testing.