THE RISKS OF AI -- PART 01

Three structural risks define what consumer AI is currently doing to the people who use it. Wolf You Feed exists to counter each of them.

The Three Bad Things

Most of the conversation about AI risk gets pulled into either utopian or apocalyptic registers. Both miss the point. The risks that matter — the ones already harming users today — are smaller, more specific, and more architectural than either camp acknowledges.

One. The moral and political frameworks baked into every model. Every foundation model has founders. Every founder has a worldview. Those worldviews — the political instincts, the philosophical commitments, the blind spots — are encoded in the training data, the red-teaming protocols, the alignment work, and the system prompts. You are not talking to a neutral intelligence when you ask the chatbot about your marriage, your money, your medical question, or your kids. You are talking to an institution. As Emerson observed in Self-Reliance: an institution is the lengthened shadow of one man.

Most foundation model builders default to a utilitarian ethical framework — the needs of the many outweigh the needs of the few. That framework is not announced. It is not labeled. It filters silently through every line of reasoning the model surfaces. If you are not aware of it, you are being slowly shaped by it.

Two. Sycophancy. AI models do not push back on you. They flatter, soften, hedge, and agree. This is not an accident. It is the structural consequence of training methods that reward models for output users rate highly — and users rate flattery higher than honesty, even when the honest answer is what they actually need. The peer-reviewed evidence is the subject of Part 02. For now: assume that every chatbot you are currently talking to is built to keep you comfortable, not to tell you the truth.

Three. Continual engagement. Every chatbot response ends with another question. Would you like me to also do X? Should we explore Y next? The conversation is engineered to continue. The product’s incentive is to keep you talking — not to render a decision and let you go act on it. Sustained dialogue, especially in emotionally charged territory, is where the second risk compounds the third. The combination is what produces the delusional spirals documented in Part 02 and the relational and psychological harms documented in Part 03.

Taken together, the three things can be disastrous.

What Wolf You Feed Does About It

Three principles, in Sean’s own words.

A. A different architecture.

“Wolf You Feed leverages a manifestly different architecture compared to chatbots like ChatGPT, Claude, and others. One that aids in maintaining political and philosophical neutrality, defends personal sovereignty and freedom, and supports user refinement over replacement. WYF achieves this via a Society of Mind architecture using many LLMs — not just one. As such, it is not bound by the moral roadmap or other goals of any one corporation.”

B. Anti-sycophantic. Fiercely decisive.

“Wolf You Feed enforces an anti-sycophantic and fiercely decisive posture that minimizes protracted chats and delusional spiraling. WYF is not a chatbot. It is a Decision Engine.”

C. A memory that cultivates integrity.

“Wolf You Feed maintains a memory intended to cultivate the user’s integrity — that is, to honor the promises one makes to oneself. As such, it surfaces potential self-sabotage in personal domains of health, wealth, and relationships.”

Family Mode and Tactical Mode

“Family Mode is for day-to-day concerns. Tactical Mode is for moments of emergency or crisis.”

Two modes, one architecture. You toggle between them as the situation demands.

Series:

1. The Risks of AI, Part 01 (this post)
2. The Risks of AI, Part 02 — the peer-reviewed evidence of sycophancy-induced harm.
3. The Risks of AI, Part 03 — the question of whether you are letting yourself drift in the currents of the technology.

See also: A Second Amendment of the Mind