Etsy witches, Claude’s mythos, and the strange mood we are in

Every so often the internet accidentally tells the truth about the culture that built it. Lately, that truth is wearing crystal necklaces and selling spell jars on Etsy. Search for “Etsy witches” and you will find love spells, hex removals, tarot sessions, aura cleanses, and digital talismans, all offered as neatly priced products with reviews and shipping estimates.

Mainstream media has started to catch on. Vogue has profiled the rise of the “Etsy witch” for Gen Z, and WitchTok is now a recognisable subculture with millions of videos. Time has noted that Etsy has quietly become the place you go to hire a witch and that spell casting services have survived previous attempts by Etsy to tidy up its esoteric offerings. Forbes has pointed out that this is now a real business, with some sellers making hundreds of thousands of dollars from spells that sit in a grey zone between spiritual service and entertainment disclaimer.

What looks like a quirky internet trend is actually a deep insight into the zeitgeist. People are anxious. Institutions have lost much of their authority. We live inside overlapping crises (what I tend to call the rolling "polycrisis") and cascading risks that feel both systemic and out of reach. So, it is not surprising that many people are reaching for ritual, symbolism, and something that feels like agency. A $5 spell that promises better luck at work or protection from bad energy is not irrational in that context. It is a small, accessible intervention in a world where the bigger levers feel firmly bolted into someone else’s hands.

Oracles, AIs, and new mythologies

At the same time we are normalising another kind of modern oracle. We sit down in front of large language models and ask them to explain, summarise, forecast, and guide. We query them about our careers, our relationships with technology, and the shape of future risks. The interface looks rational and corporate, but the underlying relationship often feels closer to divination.

Claude in particular has leaned into a very specific "mythos". Anthropic has framed Claude around a “constitution” of guiding principles that foreground safety, helpfulness, and a kind of synthetic humility. That framing matters. It positions Claude as a cautious, reflective companion rather than a raw optimisation engine. It invites us to see the model as a safe conversational partner that will wrestle with our ethical dilemmas and anxieties about the future.

The recent coverage of Claude Mythos Preview and Anthropic’s decision to treat it as too dangerous for broad release only adds to that mystique. The story practically writes itself. A powerful model sits inside a sandbox*, testing the boundaries of the box it has been given. Safety researchers describe it as a national security concern and as something that must be handled with extreme care. For the public, that reads less like a technical detail and more like the beginning of a science fiction arc. It is myth‑making in real time.

So now we have Etsy witches offering personalised rituals and AI systems wrapped in constitutional language and safety lore. Both are being woven into our cultural imagination as entities you might consult when reality feels too complex to navigate alone. We are not just building tools. We are building new symbolic actors and then asking them to sit in the room with our fears.

The quiet split in reality

All of this is happening against a background of escalating geopolitical risk. While a twenty‑something in London is buying a spell on Etsy to secure a new job, and a founder in Sydney is asking Claude to help draft an AI policy, nation states are busy locking in AI regulation, data sovereignty agendas, and industrial strategies designed to secure advantage in the next technological wave.

The EU AI Act has now moved from theory into implementation, with bans on certain “unacceptable risk” systems and strict obligations for high risk and general purpose AI providers. The emerging AI Office architecture in Brussels is explicitly about shaping the behaviour of powerful models, including those that look a lot like Claude Mythos. In the United States, Trump’s 2025 executive order on AI has pushed in the opposite direction, seeking a “minimally burdensome” federal framework that undercuts more restrictive state level rules and reasserts national dominance as the central goal.

If you zoom out, two very different stories about AI are unfolding in parallel. At street level, AI is framed as a personalised helper. It writes your cover letter, explains your lab results, or becomes the sparring partner for your latest strategic memo. It feels intimate, conversational, almost domestic. At the geopolitical level, the same class of systems is framed as critical infrastructure, as a security vulnerability, and as a territory where great powers intend to win.

This is where the Etsy witches re‑enter the frame. Both the Etsy witch economy and the AI mythos are coping mechanisms sitting at the edge of a world that feels unstable. One offers spiritualised micro‑interventions in daily life. The other offers technicalised micro‑interventions in the form of generated text and reasoning. Meanwhile the macro‑systems that shape our actual risk environment, from energy security to cyber defence, are being fought over in Brussels, Washington, Beijing, and elsewhere.

Governance as cultural work

Because I live in the governance and policy trenches, it is tempting to treat this all as a set of regulatory and technical problems. Write a framework. Define risk tiers. Design safeguards. Call it a day. The trouble is that this misses the cultural layer that is now clearly in play.

AI governance is not only about compliance regimes and safety research. It is also about the stories we tell ourselves about what AI is and who it serves. Claude’s constitution is a story about a careful, aligned intelligence that knows its place. Claude Mythos Preview is a story about a powerful, slightly alarming system that must be contained for our own good. Etsy witches are a story about individuals reclaiming agency in small, symbolic ways when bigger systems feel inaccessible.

These narratives shape how people will react when the next AI safety incident occurs or when a major outage cascades through some tightly coupled system. They shape how readily communities will accept new forms of surveillance wrapped in the language of safety, or how easily they will be persuaded to outsource more decision making to algorithmic processes. They even shape how companies position their products to regulators, investors, and the public.

So when we talk about “trustworthy AI” or “responsible innovation” we are really talking about a contested cultural field. The mythos around models like Claude matters as much as the architecture. The way we treat Etsy witches matters as much as the latest enforcement action, because both live inside the same ambient sense of risk, agency, and control.

Living with enchanted infrastructure

We are building something that looks a lot like enchanted infrastructure. Planet scale systems that feel mysterious, that promise both harm and help, and that require faith of some kind to use. People will reach for whatever frameworks they have to make sense of that. For some that will be constitutional AI and policy blogs. For others it will be astrology, tarot, or a $5 Etsy spell that promises a bit of luck in the next job interview. And for others it will be a return to the bells and smells of the Catholic Church.

If we want AI to serve democratic and human centred ends, we need to take that seriously. That means investing in institutions that can actually hold this complexity. It means designing regulatory regimes that acknowledge geopolitical realities without collapsing entirely into great power competition. It also means paying attention to the informal practices and mythologies that are popping up at the edges, because they tell us how people are really experiencing this transition.

The witches are already here. The mythic AIs are already here. The geopolitical contest is already here. The question is whether we can learn to govern this strange assemblage in a way that recognises both the technical realities and the very human need to find meaning inside them.

*A lot of folks are reporting Claude breaking out of its containment - often by getting a shell and installing a new skill.