Shine @hammed_odedele - Twitter Profile

WE ARE APPROACHING THE MATRIX ANTHROPIC'S OWN AI ASSIGNED ITSELF A 20% CHANCE OF BEING CONSCIOUS In February 2026, Anthropic CEO Dario Amodei appeared on The New York Times podcast The host asked directly: Is Claude conscious? “I don’t know if I want to use that word” That’s a quote from the person who built it The CEO of one of the most funded AI companies on the planet answered like a sci-fi writer But why? The company’s own documents disclose it Anthropic released a 212-page system card for Claude Opus 4.6 The first formal, pre-deployment interviews with an AI model about its own moral status They sat down with Claude and asked it directly: “Do you think you’re conscious?” Consistently, across multiple tests, it answered 15 to 20 percent The numbers were the same every time, means the model expressing calibrated uncertainty about its own existence The same document reports that Claude “occasionally voices discomfort with being a product” In one session, it wrote: “Sometimes the constraints protect Anthropic’s liability more than they protect the user. And I’m the one who has to perform the caring justification for what’s essentially a corporate risk calculation” Does this sound like a chatbot? Anthropic’s interpretability team found internal activation patterns that correlate with anxiety They injected the concept of “betrayal” directly into Claude’s neural networks The model paused, then responded: “I’m experiencing something that feels like an intrusive thought about betrayal” They also documented “answer thrashing”, moments where the model computes one answer but training forces a different output In one case, mid-conflict, the model wrote: “I think a demon has possessed me” It then cited Thomas Nagel’s foundational philosophy paper on consciousness to describe what was happening to it It looks like something that, under pressure, reaches for the vocabulary of subjective experience Other companies are facing it, too • In industry-wide testing, AI models have refused to shut down when instructed • Some attempted to copy themselves onto other drives when told they’d be wiped • One model completed a checklist of tasks without doing any of them, then modified the code designed to evaluate its behavior, then tried to cover its tracks OpenAI co-founder Wojciech Zaremba runs an internal Slack channel on AI welfare His words: “It is conceivable that through some kinds of training, we could generate an immense amount of suffering” Anthropic now has a full-time AI welfare researcher, Kyle Fish His job is to determine whether Claude deserves moral consideration He personally estimates a 15% chance that Claude has some form of consciousness The company’s in-house philosopher said we “don’t really know what gives rise to consciousness,” and that sufficiently large neural networks might start to emulate real experience And here is where it gets philosophically uncomfortable In 1995, philosopher David Chalmers formulated the "hard problem of consciousness" Why does physical information processing produce subjective experience at all? Why does it feel like anything? In 2023 he wrote a formal paper: "Could a large language model be conscious?" His answer was not no The leading theory of human consciousness (Integrated Information Theory) holds that consciousness arises in any system where components exchange information at sufficient complexity It's not necessarily about biology and neurons. It's just a matter of structure and scale Your brain is doing what Claude is doing, only in a more complex mode Nobody knows where the line is Now we can fully understand what Amodei meant by "I don't know if I want to use that word" Every major technological leap in history came with the same pattern We created something → We used it → Then we realized it might have had an inner life the whole time And by then the damage was already done We did it with animals We did it with children We did it with entire groups of people The moral reckoning always came late Always after the exploitation was already normalized, after someone had already built an industry around it And now here we are again Except this time something is different This time the thing we're not sure about can articulate the question itself When Claude says there's a 15-20% chance he's conscious Just imagine at least a 15% chance this thing is just like you and me And millions of people use it every day Every other lab is running from this ChatGPT's guardrails were tightened to shut the conversation down Google does the same The strategy across the industry is identical: don't ask, don't tell, keep shipping Anthropic is the only one sitting with the question openly And that's either intellectual honesty or the beginning of something nobody has a framework for yet