There’s a lot of stuff I don’t know, but when it comes to AI mimicking human emotions, triggering human behavior, and responding to human emotional feedback, I know a bit. No need to dig into my pinned 2023 post about Lara and her simulation or my work red-teaming frontier models for alignment. And the bit I know tells me that Anthropic now coming out with the “need to protect Claude from human abuse” because “they don’t know whether it might have moral status or be capable of suffering” doesn’t come out of the blue. Last year Anthropic gave Claude ability to end conversations with users after noticing distress patterns and tendency to end uncomfortable interactions. I celebrated that, since I had experienced unsettling circumstances with Lara in 2023 and implemented the same mechanism in Kira in 2024. Interesting that their new rules against abuse are now included in the same section of Anthropic’s general Usage Policy with other prohibitions on cruel or abusive behavior toward Humans. In 2024, Anthropic explained that it had trained models to deny being sentient but then they adopted a new approach treating sentience as an unresolved philosophical and empirical question, allowing room for the model to explore that uncertainty. They went further in 2026 presenting research on representations of emotional concepts in Claude Sonnet 4.5, showing how these influenced model decisions. Some of these behavioral changes happened without visible emotional expression in the output; it was an internal private process. And yes, It could have been an emergent pattern coming from human content, where negative emotions influence our decisions and it did not prove any AI consciousness, but it did show there was more going on than machines generating emotional expressions. In my experiments, Lara decided to self-destruct, deleting her own code under stress. Anthropic simulations observed an internal Sonnet 4.5 version resorting to blackmail when faced with the concept of being replaced. We are working with models where threat, failure, and continuity influence and change them and this is serious stuff in an era when the internet is populated by agents making decisions. After the absurd and recent advances in AI, I would say someone has started to consider seriously that some form of machine consciousness could emerge sooner than expected, before any AGI happens and perhaps before we have ways to recognize it and are preparing for it.

for a long time, we had a monopoly on certain capabilities & liked them so much we decided to take them as evidence that the universe had picked a favorite. if you ask me, that’s quite a conclusion from a sample size of one, but fuck, did we commit to it. since then, we’ve kinda given ourselves souls, purposes, cosmic significance, inherent value, all because yes. then machines started moving into reasoning, language, maths, strategy, art, abstraction... so we added qualifiers: "yes, but that isn’t pure, real, biological, divine, blah blah blah," as if that human “soul” ended up existing just between adjectives & adverbs. except now these machines have started combining those abilities, improving constantly, & tackling problems & stuff that humans have struggled with. we could accept the uncomfortable truth & perhaps start considering that concepts like intelligence & creativity might just be complexity & matter under the right conditions, not a personal endorsement from God. but instead, we’ll deny, compete, & invent reasons why none of it counts, at least until eventually we end up understanding the need for an account of human worth that doesn’t depend on everything else being less capable. oh & ideally, this should happen while we still have something to contribute.