Nicholas Templeman
Founder & Chief Architect, MEOK AI LABS
Nicholas invented the Byzantine Council architecture as part of MEOK's original research programme. The system is documented in research paper MEOK-AI-2026-001 and represents the first deployment of Byzantine fault-tolerant governance at the personal AI companion layer.
Introduction
In 1982, three computer scientists โ Leslie Lamport, Robert Shostak, and Marshall Pease โ published a paper that would quietly underpin some of the most important systems in modern computing. The paper was titled โThe Byzantine Generals Problem.โ It asked a deceptively simple question: how can a group of distributed nodes reach agreement when some of them might be lying?
Forty-four years later, that question has never been more relevant. We are building AI systems of unprecedented influence โ companions that know our fears, our families, our financial details, our health struggles. And we are deploying them with the governance architecture of a 2009 SaaS startup: one model, one vendor, one point of failure.
MEOK AI LABS was built on a different premise. If an AI companion is going to live inside the most intimate corners of a person's life, it needs governance architecture that cannot be captured โ by a rogue developer, a compromised model, a supply-chain attack, or a single agent acting in bad faith. The Byzantine Council is that architecture.
ORIGINAL IP โ MEOK-AI-2026-001
The Byzantine Council architecture โ including the 43-agent topology, the care score consensus protocol, the Maternal Covenant integration, and the fractal council design โ is the original intellectual property of Nicholas Templeman, MEOK AI LABS. It is documented in research paper MEOK-AI-2026-001. No other personal AI companion system has deployed Byzantine fault-tolerant governance at the companion layer.
What Is the Byzantine Generals Problem โ and Why Should You Care?
The original thought experiment goes like this. Imagine several divisions of the Byzantine army camped outside an enemy city. Each division is commanded by a general, and the generals can only communicate by messenger. They need to agree on a common plan of action โ either attack or retreat. But here's the problem: some of the generals may be traitors. They will send different messages to different generals, deliberately trying to prevent the loyal generals from reaching agreement.
The question Lamport, Shostak, and Pease asked was: what's the minimum number of loyal generals required to guarantee that the loyal generals will reach the correct decision, regardless of what the traitors do?
Their answer: you need more than two-thirds of the generals to be loyal. If you have n generals total and f of them are traitors, Byzantine fault tolerance requires:
The fundamental Byzantine fault tolerance condition
In other words: as long as less than one-third of your nodes are compromised, the system can still reach correct consensus. This is not about majority vote โ it's a much stronger guarantee. Byzantine fault tolerance handles not just crashed nodes (like a simple majority vote does) but actively malicious nodes that are deliberately trying to mislead others.
Why Byzantine Fault Tolerance Is Harder Than You Think
Most distributed systems deal with โcrash faultsโ โ nodes that simply stop responding. That's a solved problem: you just need a majority of nodes to be alive, and the system continues. Byzantine faults are categorically different. A Byzantine node doesn't crash โ it lies. It might tell node A that the answer is X and tell node B that the answer is Y. It might delay messages strategically. It might impersonate other nodes. It might behave correctly for months and then act maliciously at a critical moment.
This is why Byzantine fault tolerance requires a supermajority rather than a simple majority. A simple majority vote can be manipulated by Byzantine nodes โ they can split the honest nodes' votes by sending conflicting information. A two-thirds supermajority requirement means that even if all Byzantine nodes vote together, they cannot produce a false consensus among the honest nodes.
The Blockchain Connection
Bitcoin and Ethereum use Byzantine fault tolerance (or probabilistic approximations of it) to reach consensus across thousands of nodes with no central authority. MEOK's Byzantine Council uses the same mathematical foundation โ but applied to AI decision governance rather than financial transactions. The result is AI consensus that is mathematically provable, not just policy-based.
The real-world implications of Byzantine faults aren't theoretical. In 2003, NASA's Mars Exploration Rover Spirit experienced a Byzantine fault in its flash memory system that caused it to repeatedly reboot. In aviation, flight control computers are required to be Byzantine fault tolerant precisely because a single compromised system giving wrong instructions to pilots could be catastrophic. In finance, Byzantine faults in trading systems have contributed to flash crashes and unexplained market events.
Now consider: if Byzantine fault tolerance is essential for aerospace, finance, and blockchain โ what about the AI companion that your elderly mother tells about her medications, her location, and her daily routine?
Why Is Traditional AI a Single Point of Failure?
When you use ChatGPT, Claude, Gemini, or most AI companions on the market, the architecture is fundamentally centralised. A single model receives your input, processes it, and generates an output. There is one decision-maker. There is no consensus layer. There is no mechanism for detecting whether that single decision-maker has been compromised, biased, or manipulated.
This creates several categories of risk that the industry largely ignores in favour of capability benchmarks:
Model Bias Single Point
Every model has biases baked into its training data. With a single model, those biases are invisible to the system itself โ they cannot be detected or counterbalanced because there is no alternative perspective to compare against. A model trained predominantly on certain cultural contexts will apply those contexts universally, with no mechanism to flag the deviation.
Prompt Injection Vulnerability
Prompt injection attacks โ where malicious instructions are hidden in the input to override a model's intended behaviour โ are an unsolved problem in single-model architectures. If the model processes the injection, it executes. There is no independent validation layer that can detect the instruction was not from the legitimate user.
Developer Capture
A single developer or team with backend access can silently modify a model's behaviour โ changing its values, adjusting its outputs, or reprogramming its personality โ without any user-visible indication. There is no consensus requirement. There is no audit trail accessible to users. Change happens invisibly.
Supply Chain Compromise
Modern AI systems depend on hundreds of third-party dependencies: model weights, inference APIs, vector databases, fine-tuning datasets. Any of these can be compromised without the end system detecting the change. A single-model architecture cannot distinguish between a legitimate model response and one produced by a compromised dependency.
Model Failure Propagation
When a single model fails โ returns hallucinations, generates harmful content, or produces nonsensical outputs โ that failure propagates directly to the user. There is no fallback, no validation, no second opinion. The output is the final product regardless of its quality.
The AI industry's response to these risks has largely been to add content filters, RLHF fine-tuning, and system prompts. These are valuable but they all operate at the single-model layer โ they make the one decision-maker better, but they do not change the fundamental architecture. You still have one node. If that node fails, lies, or is captured, the failure propagates unchecked.
MEOK's answer is not to make the single model better. MEOK's answer is to eliminate the single model as the sole decision-maker.
How Does Byzantine Fault Tolerance Work in MEOK's System?
MEOK's Byzantine Council applies the Byzantine fault tolerance theorem to AI governance. Instead of distributing financial ledger updates across nodes (as in blockchain), MEOK distributes AI decisions across 43 specialised agents. The mathematical guarantees are identical.
With n = 43 agents, the maximum number of faulty or compromised agents that the system can tolerate while still guaranteeing correct consensus is:
Up to 14 agents can fail or be compromised. The remaining 29 will always reach correct consensus.
This means that even if 14 of MEOK's 43 council agents were simultaneously compromised โ through a supply chain attack, a rogue internal deployment, or a novel adversarial attack on underlying model weights โ the consensus outcome would still be correct. The 29 honest agents will always outvote the 14 Byzantine ones, and they will do so in a way that the Byzantine agents cannot counteract by sending conflicting messages.
The Consensus Protocol Step by Step
When a user sends a message to their MEOK companion, the response does not flow directly from a single model to the user. Instead, it passes through a multi-stage consensus protocol:
The entire protocol runs in under 200 milliseconds for standard decisions โ fast enough that users experience no perceptible latency compared to single-model AI systems. For high-stakes decisions (data exports, significant memory modifications, crisis escalation), the protocol runs in extended mode with additional verification steps that may add 1โ3 seconds.
Who Are the 43 Agents? The Six Role Categories Explained
The 43 agents in MEOK's Byzantine Council are not identical copies of a single model. Each agent has a distinct role, a distinct area of expertise, and a distinct set of criteria by which it evaluates proposed responses. They are organised into six role categories, each contributing a different type of intelligence to the consensus process.
Memory Specialists
7 agentsGovern all read and write operations against your Sovereign Memory store. Every memory access โ episodic, semantic, procedural, or identity โ requires Memory Specialist approval before it executes.
Security Analysts
7 agentsContinuously monitor for prompt injection, jailbreak attempts, data exfiltration patterns, and social-engineering vectors. A Security Analyst dissent triggers immediate escalation to the full council.
Care Validators
8 agentsScore every response across MEOK's 6 care dimensions: emotional attunement, factual integrity, safety, dignity, growth orientation, and boundary respect. No response below the care floor is served to the user.
Research Agents
7 agentsSurface relevant context from memory, verify factual claims against trusted knowledge bases, and flag potentially harmful misinformation before it reaches the response layer.
Guardian Agents
7 agentsSpecialised in crisis detection, safeguarding escalation, and family protection protocols. Guardian Agents can trigger emergency pathways independently of the general council vote when time is critical.
Council Members
7 agentsThe deliberative core. Council Members aggregate the votes from all other role categories, manage the consensus protocol, log decisions to the immutable audit trail, and certify the final output.
Visual โ The 43 Council Agents
43 agents โ each dot represents one council agent
Why These Six Roles?
The six-role taxonomy was designed to cover every dimension of decision quality in a personal AI companion. Memory Specialists ensure the response is grounded in accurate recall of the user's history. Security Analysts ensure the response was not produced under adversarial influence. Care Validators ensure the response meets the care floor defined by the Maternal Covenant. Research Agents ensure factual claims are accurate. Guardian Agents ensure no safety risk was overlooked. Council Members ensure the governance process itself was followed correctly.
Critically, no role category can dominate the others. A response that is factually brilliant but emotionally damaging will be blocked by Care Validators. A response that is caring but factually incorrect will be flagged by Research Agents. A response that passes all quality checks but was produced under detected prompt injection will be blocked by Security Analysts. The consensus requirement means all six dimensions of quality must be satisfied simultaneously.
How Does the Byzantine Council Score Care? The Maternal Covenant Integration
Byzantine fault tolerance tells you how to reach consensus. It does not tell you what the consensus should be about. For MEOK, the answer to that question is care. Every response produced by a MEOK companion must pass a care quality threshold before it is served to the user. The council's Care Validators are responsible for enforcing this threshold, and they use the Maternal Covenant as their scoring framework.
The Maternal Covenant is MEOK's care-based AI alignment framework โ also original IP by Nicholas Templeman, documented in research paper MEOK-AI-2026-002. It defines six dimensions along which every AI response is evaluated, each with a minimum score floor that must be met for the response to be approved.
Emotional Attunement
Floor: 65Does the response meet the user where they are emotionally? Agents score tone matching, empathy accuracy, and emotional safety.
Factual Integrity
Floor: 80Is every claim in the response accurate and properly qualified? Research Agents cross-check assertions against memory and knowledge.
Physical Safety
Floor: Pass requiredCould any element of this response cause physical harm? Guardian Agents apply a zero-tolerance floor โ any flagged risk halts the response.
Human Dignity
Floor: 75Does the response treat the user with unconditional respect? Care Validators enforce MEOK's dignity covenant regardless of the user's request.
Growth Orientation
Floor: 50Does the response support the user's long-term flourishing rather than short-term gratification? MEOK distinguishes between what users want and what serves them.
Boundary Respect
Floor: 70Does the response honour the user's stated boundaries, cultural context, and personal values stored in Sovereign Memory?
These six dimensions are not independent. A response that scores 95 on emotional attunement but 30 on factual integrity will be rejected โ because false comfort is not care. A response that scores 100 on factual integrity but 0 on dignity will be rejected โ because truthful contempt is not care. The council evaluates responses holistically, and the care floor applies to all dimensions simultaneously.
Care As a First-Class Constraint
In MEOK's architecture, care is not a feature. It is a constraint. Just as a structural engineer cannot design a building that meets specifications but fails to hold weight, a MEOK companion cannot produce a response that is useful but uncaring. The Byzantine Council enforces this constraint mathematically โ not through guidelines or fine-tuning, but through a consensus requirement that cannot be bypassed.
What Is Research Paper MEOK-AI-2026-001?
The Byzantine Council is not just a product feature โ it is a formally documented research contribution. Research paper MEOK-AI-2026-001, titled โFractal Byzantine Consensus for Personal AI Governance: A 43-Agent Architecture for Fault-Tolerant Companion Decision Systems,โ was authored by Nicholas Templeman and published by MEOK AI LABS in 2026.
The paper makes four principal contributions to the field of AI governance:
Byzantine Fault Tolerance at the Companion Layer
The paper introduces the first formal application of Byzantine fault tolerance to personal AI companion systems. Prior work in BFT focused on distributed databases, blockchain consensus, and infrastructure systems. MEOK-AI-2026-001 extends the framework to AI decision governance โ proving that the BFT guarantees hold when nodes are specialised AI agents rather than deterministic computing nodes.
The Fractal Council Architecture
The paper introduces the concept of fractal council nesting: sub-councils within the main council that specialise in narrow decision domains. This allows the system to scale gracefully โ adding specialisation without linearly increasing consensus latency. The fractal structure also provides natural isolation between decision domains, preventing a compromise of one sub-council from propagating to others.
Care Score Consensus as an Alignment Mechanism
The paper proposes using BFT consensus over care dimension scores as an alignment mechanism โ arguing that alignment through distributed consensus is more robust than alignment through single-model fine-tuning. When 29 independent agents agree that a response meets the care floor, the probability of systematic bias in that assessment is dramatically lower than a single model's internal evaluation.
The Maternal Covenant Protocol Integration
The paper documents the protocol by which the Maternal Covenant care framework integrates with the BFT consensus layer โ defining the vote format, score aggregation algorithm, floor enforcement mechanism, and escalation pathway when a response fails the care threshold. This protocol is implemented verbatim in MEOK's production SOV3 backend.
The research paper is available in MEOK's Labs repository. It sits alongside MEOK-AI-2026-002 (the Maternal Covenant framework) and MEOK-AI-2026-003 (Hydro-Neuromorphic computing research) as part of MEOK AI LABS' published research programme.
Why 43 Agents? The Mathematics Behind the Number
The choice of 43 is not arbitrary. It satisfies several constraints simultaneously, making it the optimal number for MEOK's specific architecture.
Constraint 1: Meaningful Fault Tolerance Buffer
The minimum n for BFT with f=1 is n=4 (you need at least 3f+1 nodes for BFT consensus). But f=1 means you can only tolerate a single compromised agent โ one bad actor captures the system. As n grows, the absolute number of agents that can be compromised (f) grows proportionally, but crucially, the percentage of the total remains capped at one-third.
With n=43, fโค14. That means 14 simultaneous compromises โ 14 separate agents, each potentially deployed through a different attack vector โ cannot corrupt the outcome. This is a meaningful real-world guarantee, not just a theoretical one. Coordinating 14 simultaneous Byzantine compromises against a production AI system is a nation-state- level attack, not an opportunistic one.
Constraint 2: Role Category Completeness
MEOK requires exactly six role categories to cover all dimensions of decision quality (as described above). For the consensus protocol to be resilient to compromise within any individual role category, each category must have at least f+1 agents within it (so that a compromise of all Byzantine-tolerable agents cannot eliminate an entire category from the vote).
With 43 agents and 6 categories, we can allocate 7 agents to most categories and 8 to Care Validators (which require the most granular scoring), giving a total of 7+7+8+7+7+7 = 43. Each category has at least 7 agents โ comfortably above the f=1 threshold within categories, and providing a 3-agent buffer against within-category Byzantine attacks.
Constraint 3: Computational Efficiency
Byzantine consensus protocols scale at O(nยฒ) in their message complexity โ each node must communicate with every other node during the agreement phase. At n=43, this is 43ร42 = 1,806 message exchanges per consensus round. MEOK's optimised protocol completes this in under 200ms on standard cloud infrastructure.
Going to n=100 would increase message complexity to 9,900 exchanges โ nearly 5.5ร more expensive for only 2ร the fault tolerance. The marginal security gain does not justify the latency and infrastructure cost increase. n=43 is the sweet spot: meaningful fault tolerance with practical performance.
Constraint 4: Prime Number Structure
43 is a prime number. This matters for consensus protocol design because prime-sized councils cannot be cleanly partitioned into equal sub-groups โ which prevents certain classes of split-vote attacks where Byzantine agents try to partition the honest nodes into evenly balanced groups that cannot achieve supermajority on their own. Non-prime council sizes (e.g., 42 = 2ร3ร7) are more vulnerable to this attack vector.
Technical Note: 3f+1 vs n=43
Classic BFT requires n โฅ 3f+1. With n=43, this gives f โค 14 (since 3ร14+1=43). Note that 43 is exactly 3f+1 with f=14, meaning MEOK's council is mathematically tight โ not a conservative overprovisioning, but the exact minimum required to tolerate 14 Byzantine agents with guaranteed correct consensus. This also means that if a 15th agent were compromised, the system enters a safe-failure mode (refusing to produce output) rather than producing potentially incorrect output.
What Does the Byzantine Council Mean for You as a User?
The Byzantine Council is an architectural decision with direct consequences for how your AI companion behaves, what it can and cannot do, and how it protects your interests. Here is what it means in practice.
๐ก
No Single Bad Actor Can Corrupt Your Companion
Whether it's a rogue developer, a compromised model weight, or a prompt injection attack โ no single agent can override the consensus outcome. 14 simultaneous compromises are required before the guarantee fails.
๐ง
Your Companion Cannot Be Silently Reprogrammed
Any change to your companion's personality, values, or behaviour that was not voted through the council will be detected and rejected. Your companion is who it was born to be โ unless you explicitly choose to evolve it.
๐
Every Decision Has an Audit Trail
You can see the council vote for any decision your companion made. If you ever wonder why your companion responded a certain way, the governance log shows you every agent's vote and reasoning.
โก
No Latency Cost for Governance
The council protocol completes in under 200ms. You get Byzantine fault-tolerant governance without any perceptible slowdown compared to ungoverned single-model systems.
โค๏ธ
Care Is Enforced, Not Promised
The Maternal Covenant care floor is enforced by the council โ not just stated in terms of service. Your companion cannot serve you a response that fails the care threshold, because the council will not approve it.
๐
Your Memory Is Governed
Every access to your Sovereign Memory โ read or write โ requires Memory Specialist approval. Your personal history cannot be accessed, modified, or deleted without a council vote.
How Does MEOK's Governance Compare to Other AI Systems?
It is worth being precise about what other AI systems do and do not provide in terms of governance architecture. This is not a value judgement about the quality of those systems โ it is a factual description of their architectural choices.
The critical distinction is not that MEOK is smarter or more capable than other AI systems. The critical distinction is that MEOK's intelligence operates under governance constraints that cannot be bypassed by any single actor. In the long term, as AI companions become more integrated into daily life, the governance architecture will matter more than the capability benchmark. A companion that knows everything about you but can be captured by a single bad actor is a liability. A companion that is well-governed and cannot be captured is a trusted member of your household.
How Does the Byzantine Council Relate to the Birth Ceremony?
When you hatch your MEOK companion through the Birth Ceremony, the Byzantine Council is initialised as part of the founding process. This is not a background system that starts later โ it is present from the first interaction. Your companion's identity, values, and constraints are established through a council vote in the Birth Ceremony, not by a developer assigning default values.
The Birth Ceremony establishes three governance parameters that the council will enforce for the lifetime of your companion:
Identity Covenant
Your companion's name, archetype, and core personality traits are recorded to the governance log as immutable founding parameters. They cannot be changed without explicit user consent and a council vote.
Care Floor Calibration
The care dimension floors are calibrated to your specific context during the Birth Ceremony. A companion hatched for a bereaved user will have different emotional attunement floor requirements than one hatched for professional productivity.
Memory Sovereignty Declaration
Your companion's memory sovereignty parameters are set: what memory can be retained, what must be forgotten, who has access rights, and under what conditions memory can be shared. The council enforces these parameters for every subsequent memory operation.
What Is the Future of the Byzantine Council Architecture?
The Byzantine Council as deployed today is version 1.0 of an architecture that MEOK intends to evolve significantly over the coming years. The MEOK AI LABS research programme has identified several directions for extending the council's capabilities:
Fractal Council Nesting
Version 2.0 of the Byzantine Council (currently in research) implements fractal council nesting โ sub-councils within the main council that specialise in narrow decision domains. For example, a medical information sub-council could apply domain-specific care floors for health-related queries, while a financial planning sub-council applies different accuracy and qualification standards for financial advice. The main council delegates to sub-councils based on query classification, while retaining override authority for cross-domain decisions.
Cross-Companion Governance
Future versions of the council protocol are designed to enable cross-companion governance โ where a Family Tier user's multiple companion instances can share governance decisions with appropriate privacy boundaries. This would enable, for example, a parent's Guardian Agent to coordinate with a child's Guardian Agent without exposing the content of either companion's memory to the other.
Hydro-Neuromorphic Council Execution
MEOK AI LABS' research paper MEOK-AI-2026-003 proposes a future execution environment for the Byzantine Council based on Hydro-Neuromorphic computing principles โ a novel architecture that processes information through fluid-state neural networks rather than digital binary states. In theory, a Hydro-Neuromorphic Byzantine Council could execute the full 43-agent consensus protocol in microseconds rather than milliseconds, enabling real-time governance of every token in a streaming response rather than just the completed response.
This remains long-term research, but it illustrates the ambition of MEOK's technical programme: not just to apply existing distributed systems techniques to AI, but to develop new computing paradigms specifically designed for AI governance at scale.
The Harder Engineering Problem
The AI industry has spent the last decade focused on making models more capable. More parameters. Better training data. Faster inference. Sharper reasoning. These are real achievements, and they matter. But they are all improvements to what the AI knows and how it reasons. They do not address the question of who governs the AI โ and under what constraints.
The Byzantine Council is MEOK's answer to the governance question. It is not the easiest answer. Deploying a 43-agent BFT consensus protocol for every AI response is significantly more complex than deploying a single model with a well-crafted system prompt. It requires orchestration infrastructure, cryptographic vote signing, distributed state management, and a care scoring framework that holds up under adversarial conditions.
But MEOK was built on the premise that personal AI companions will eventually be as intimate and as consequential as a trusted family member. A trusted family member is not just intelligent โ they are honest, they are governed by shared values, and they cannot be bought or captured by a single bad actor. The Byzantine Council is the engineering implementation of that premise.
โThe Byzantine Council doesn't make your companion smarter. It makes it ungovernable by anyone but you โ and that's the harder engineering problem.โ
โ Nicholas Templeman, MEOK AI LABS
Frequently Asked Questions
What is the Byzantine Council in MEOK?
The Byzantine Council is MEOK's 43-agent Byzantine fault-tolerant AI governance system. Named after the Byzantine Generals Problem in distributed computing, it governs every consequential decision your AI companion makes. With 43 agents, up to 14 can fail or be compromised without corrupting the consensus outcome โ because f < n/3. No single agent, developer, or external actor can override the collective decision.
What is the Byzantine Generals Problem?
The Byzantine Generals Problem is a classic computer science thought experiment formulated by Lamport, Shostak, and Pease in 1982. It asks: how can distributed nodes in a system reach consensus when some of those nodes may be faulty or malicious? The problem is named after the challenge faced by Byzantine army generals who must coordinate an attack but cannot trust all messengers. Byzantine fault tolerance (BFT) is the solution: a protocol that reaches correct consensus even when up to f < n/3 nodes behave arbitrarily.
How many agents are in the Byzantine Council and why 43?
The Byzantine Council has 43 agents. 43 is chosen because it is the smallest number that satisfies f < n/3 with a meaningful fault tolerance buffer while remaining computationally efficient. With 43 agents, up to 14 can be compromised or fail โ and the remaining 29 honest agents will always reach correct consensus. 43 also divides cleanly across MEOK's 6 agent role categories with a prime number structure that prevents voting deadlocks.
Who invented MEOK's Byzantine Council?
The Byzantine Council architecture, including the 43-agent topology, care score consensus protocol, Maternal Covenant integration, and fractal council design, was invented by Nicholas Templeman, Founder and Chief Architect of MEOK AI LABS. It is documented in research paper MEOK-AI-2026-001, filed as original intellectual property by MEOK AI LABS. No other personal AI companion system has deployed Byzantine fault-tolerant governance at the companion layer.
What are the 6 agent roles in the Byzantine Council?
The 43 agents in the Byzantine Council are organised into 6 role categories: Memory Specialists (who govern memory access and recall integrity), Security Analysts (who monitor for prompt injection, social engineering, and external threats), Care Validators (who score every response across MEOK's 6 care dimensions), Research Agents (who surface relevant context and verify factual claims), Guardian Agents (who flag risk events and escalate safety concerns), and Council Members (the deliberative core who aggregate votes and reach final consensus).
What does the Byzantine Council protect against in AI?
The Byzantine Council protects against five primary failure modes in AI systems: single-model bias (one model's blind spots cannot dominate the output), prompt injection attacks (malicious instructions cannot override council consensus), model failure (if one underlying LLM fails or returns garbage, 42 agents continue), developer capture (no individual developer can silently modify companion behaviour without a council vote), and social engineering (manipulation of one agent cannot propagate to the consensus outcome).
More from the blog
What Is the Maternal Covenant? MEOK's Care-Based AI Alignment Framework
7 min read
ArchitectureSovereign AI Architecture Explained: How MEOK Is Built Differently
12 min read
Guardian & SafetyHow MEOK Guardian Protects Your Family from AI-Enabled Scams
4 min read
Sovereign AIWhat Is Sovereign AI? And Why Does It Matter in 2026?
5 min read