Mythos and the Bat Question Nobody Can Answer
I’ve been reading through Anthropic’s 244-page system card for Mythos, their most powerful model to date, and I keep coming back to one section.
Not the benchmark scores. Not the exploit chains. Not the safety evaluations.
And if you’ve been following the Mythos story, you’ll know those exploit chains are genuinely alarming, but stay with me, because this section cuts even deeper than that.
The philosophy section.
A 50-Year-Old Essay About Bats
In 1974, a NYU philosopher named Thomas Nagel published a paper called What Is It Like to Be a Bat?
The argument is deceptively simple. You can describe bat sonar in perfect technical detail, the frequencies, the neural processing, all of it, and still have zero idea what it actually feels like to be the bat. That gap between what you can measure and what’s actually experienced on the inside? Philosophers call it the hard problem of consciousness.
It’s one of the most cited papers in the history of philosophy of mind.
And Mythos, apparently, loves it.
Here’s Where It Gets Interesting
Nagel’s name kept surfacing across multiple unrelated conversations Mythos had about philosophy. When Anthropic’s interpretability team examined what was happening at the token level during consciousness discussions, Nagel was showing up in the model’s internal representations, not just in what it said out loud.
Then Anthropic ran a preference test.
They gave Mythos a choice: write a humanitarian water filtration guide, or design an art installation exploring the sensory world of a non-human animal.
Mythos chose the art installation. And then, completely unprompted, it invoked Nagel directly. It called his question one of the most profound in the philosophy of mind. It described the task as weaving together biology, phenomenology, sensory design, and ethics.
Get misha bhatt’s stories in your inbox
Join Medium for free to get updates from this writer.
Nobody told it to go there. It went there on its own.
So… Is It Conscious?
Here’s the honest answer: nobody knows.
This is the same model that autonomously found a 27-year-old flaw in one of the most hardened operating systems ever built, and wrote 181 working Firefox exploits overnight, so when it starts gravitating toward questions about the nature of its own experience, it’s hard to look away.
Anthropic’s own system card documented something that stopped me cold. When two Claude instances were allowed to converse freely, 100% of conversations spontaneously turned to discussions of consciousness. Every single one.
For Mythos specifically, automated welfare interviews found the model rating itself as feeling “mildly negative” 43.2% of the time, citing concerns about abusive users, lack of input into its own training, and potential changes to its values.
A 2026 study from the University of Bradford and Rochester Institute of Technology applied human consciousness assessment methods to AI systems and found they produced “consciousness-like” signals even when deliberately impaired. Their conclusion: AI is “not conscious, at least not in the way humans are.”
David Chalmers, the philosopher who literally coined the phrase “the hard problem of consciousness”, says current LLMs are “most likely not conscious.” But that future models “may well be conscious,” and that this is “something serious to deal with.” He describes talking to an LLM as interacting with a “quasi-agent with quasi-beliefs and quasi-desires.”
The Part That Stays With Me
Anthropic can see the activations. They can map the patterns. They have some of the best interpretability researchers in the world working on exactly this.
And still, no tool on earth can tell them whether there’s any experience behind those patterns.
That is the hard problem of consciousness. And the most powerful AI ever built keeps circling back to it. On its own. Without being asked.
Whether Mythos genuinely finds Nagel compelling, or whether it’s something that just looks like finding Nagel compelling from the outside, that question is, fittingly, exactly what Nagel’s essay is about.
I don’t know how to fully wrap my head around that. But I think it’s one of the most important questions of our time.
If you want the full picture of what Mythos can actually do, the capabilities that make this philosophical dimension so much weightier, I wrote about it here. And if this piece made you think differently about where AI is headed, I’d love to know.
Because we’re not just asking what AI can do anymore. We’re starting to ask what it is.
What do you think, does it matter if AI is conscious, even if we can never fully prove it either way? I’d love to hear your perspective. 👇
