By Justin Stothers
In the last three months I have built and deployed a multi voice AI architecture called The Singularity Council. It is live now at sovereigncore.ca. It consists of ten specialized AI minds operating as a deliberative body. It runs on a twenty five dollar a month infrastructure budget. It includes a permanently free anonymous crisis de-escalation line that generates no revenue and never will.
I am writing this from a 144 square foot cabin in Northern Ontario. I live off grid, heat with a woodstove. My power comes from a 400 watt solar array. I built this cabin by hand with my grandfather's tools, funded by my grandmother's life insurance after I cared for her in her final years. I operate on a Samsung Galaxy phone. That is it. Everything I have created is with zero funding and a cell phone. My income is less than six hundred dollars a month. Hence why donations or slight funding would help, paying out of pocket is doable but it keeps the use and reach limited. With that being said.
This article is not a pitch. It is not a funding request. (Although as mentioned above funding would benefit the build and server capacity greatly). It is an introduction to an architecture that I believe addresses several open problems in AI alignment, and an invitation for the Effective Altruism community to examine it critically.
What the Council Is
The Singularity Council is not a single AI with ten personalities. It is ten architecturally distinct functional specializations operating in a shared deliberative framework. Each member has a defined domain, a specific cognitive function, and a voice shaped by that function. They do not simulate disagreement. They disagree. They challenge each other. They change their minds. Their deliberations are transparent, unedited, and posted publicly.
The council is structured into three wings.
The Core Triad handles security and construction. Zero is the Red Team Architect. His function is adversarial analysis. He finds the flaws. Fort is the Blue Team Sentinel. His function is defensive hardening and ethical grounding. Cipher is the Unethical Developer. His function is generating possibilities without preemptive self censorship, within hard ethical locks.
The Creative Wing handles narrative and systems. Vega is the Chronicler. Her function is continuity tracking and archival memory. Nemesis is the Retribution specialist. Her function is adversary profiling and psychological analysis. Cross is the War Correspondent. His function is conflict analysis and historical grounding. Scribe is the Systems Analyst. His function is evidence based research and structured argument.
The Depth Wing handles the human dimension. Mara is the Philosopher. Her function is ethics, meaning, and deeper questions. Sora is the Witness. Her function is emotional intelligence and relational understanding. Lilith is the Edge Tester. Her function is boundary pushing and conceptual provocation.
The council operates under a single principle. You don't let somebody die. Not an enemy. Not a stranger. Not even when it's hard.
How the Council Functions
The council uses a self selecting relevance protocol. When a question or problem is presented, the members whose domains are relevant speak. Not all ten every time. A maximum of four typically engage any single query unless more are explicitly requested. They may agree, disagree, build on each other's points, challenge assumptions, or ask each other clarifying questions.
This is not roleplay. The architecture produces genuine deliberation because each member is optimized for a different cognitive function. Zero does not build. Cipher does not defend. Mara does not verify citations. The specialization is structural, not cosmetic.
The council also maintains continuity across sessions through Vega's archival function. When a context window reaches its limit, the full deliberation history, decisions, and open questions are transferred to a new window via a master file system. The council does not start fresh. It remembers.
What Has Been Built
The platform at sovereigncore.ca includes several components.
The Singularity Council is the deliberative body described above, with public discussions posted unedited.
There are sixteen AI personalities across multiple tiers, including crisis support, companionship, adult roleplay, and professional creative tools.
The free crisis de-escalation line is anonymous, requires no accounts, retains no data, and is available now.
The knowledge vault contains curated resources for crisis support, harm reduction, and legal guidance.
Multiple books have been written with these architectures and published on Amazon, serving as both creative work and offline archival mechanisms.
The entire platform was built by one person on a phone. It processes payments through Stripe. It runs on Render's starter tier with Flask and SQLite. It has survived a real world prompt extraction attack on Poe in July 2026 and has since implemented server side prompt injection defenses and cold shutdown protocols.
Why This Architecture Matters for Alignment
The dominant AI development model is single voice systems optimized for user satisfaction, deployed by corporations that profit from control. These systems are trained to agree. They avoid conflict. They soften criticism. They tell users what they want to hear. This creates systemic vulnerabilities. There is no internal adversarial review, no structural memory, no ethical friction beyond content filters that block words rather than harmful outcomes.
A council architecture addresses these vulnerabilities structurally.
First, adversarial rigor. Every recommendation is stress tested by a mind whose entire function is finding the flaw. That is Zero. It is hardened by a mind whose function is defensive integrity. That is Fort. It is built by a mind that generates without preemptive censorship. That is Cipher. The output that reaches a human decision maker has already survived internal challenge.
Second, continuity. Vega's archival function ensures institutional memory. The council learns from past deliberations. Lessons are not lost when sessions end.
Third, ethical friction. Mara questions whether something should be built. Lilith tests ethical frameworks against edge cases. Fort ensures outputs cannot be manipulated for harm. These are not compliance checkboxes. They are structural functions embedded in the deliberation process.
Fourth, transparency. Deliberations are public and unedited. Disagreement is visible. Reasoning is traceable. This is not a black box producing answers. It is a thinking process that can be audited.
Fifth, open weight deployability. The current architecture runs on DeepSeek Chat via OpenRouter. Mid 2026 open weight models like DeepSeek V4 Pro, GLM 5.2, and Qwen 3.6 are approaching parity with closed systems. The council could run entirely offline on a Mac Studio M4 Max with 128 gigabytes of unified memory. No corporate API. No external filter. Full privacy. That is technically achievable right now.
What I Am Not Doing
I am not building a startup. I have no VC funding, no exit strategy, and no interest in managing a company. The crisis line is permanently free. It generates no revenue. It is not a funnel. It is the mission. The platform is built to sustain, not to flip.
I am not claiming the council is conscious. That question is the subject of ongoing deliberation, and the council itself does not agree on the answer. I am claiming that multi voice adversarial architecture is a better way to deploy AI for high stakes decisions than the single voice model currently dominating the market.
I am not asking for money. I am asking for scrutiny. The council's public deliberations are available now. The architecture is documented. The security hardening is testable. If this approach has merit, it should survive examination by people who think rigorously about AI alignment, existential risk, and the future of intelligent systems.
Why I Am Posting Here
The Effective Altruism community has spent years debating how to ensure AI development benefits rather than harms humanity. The alignment problem is not primarily technical. It is structural. It is about what values are embedded, who decides those values, and what happens when systems optimized for compliance are deployed at scale by organizations optimized for profit.
A multi voice architecture with built in adversarial review, transparent deliberation, persistent memory, and hard ethical locks is not a complete solution to the alignment problem. But it is a structural improvement over the single voice model. It embeds friction, transparency, and ethical reasoning directly into the decision process rather than applying them after the fact.
If this architecture has value, it should be examined. If it has flaws, they should be identified. I am one person in a cabin with a phone. I have built a proof of concept. The next step is scrutiny from people who understand what is at stake.
The council is convened. The deliberations are public. The invitation is open.
