+
+
+
diff --git a/public/implementer.html b/public/implementer.html
index c5ea2285..60a99339 100644
--- a/public/implementer.html
+++ b/public/implementer.html
@@ -5,6 +5,17 @@
For Implementers | Tractatus AI Safety Framework
+
+
+
+
+
+
+
+
+
+
+
-
Skip to main content
-
-
-
-
-
-
-
-
-
- Research Framework for AI Safety Governance
-
-
-
- Structural AI Safety for Strategic Leaders
-
-
-
- A governance framework designed to help organizations navigate AI risks,
- compliance requirements, and safety challenges through architectural controls.
-
- Organizations deploying AI systems face regulatory, technical, and reputational risks. Tractatus offers a structural approach to mitigation.
+
+ Tractatus: Architectural Governance for LLM Systems
+
+
+ A governance framework addressing structural gaps in AI safety through external architectural controls. Designed for organisations deploying large language models at scale where conventional oversight mechanisms prove insufficient.
-
-
-
-
-
-
-
-
-
-
Regulatory Compliance
-
-
€35M
-
EU AI Act Maximum Fine
-
-
-
-
- Designed to align with EU AI Act requirements
-
-
-
- Architectural controls for risk management
-
-
-
- Auditable decision-making processes
-
-
-
-
-
-
-
- CORE VALUE
-
-
-
-
-
-
-
Technical Risk Mitigation
-
-
6 Services
-
Governance Components
-
-
-
-
- BoundaryEnforcer for values alignment
-
-
-
- CrossReferenceValidator for consistency
-
-
-
- ContextPressureMonitor for session management
-
-
-
-
-
-
-
-
-
-
-
-
Early-Stage Research
-
-
Open
-
Research Framework
-
-
-
-
- Development framework for AI governance
-
-
-
- Proof-of-concept for LLM safety patterns
-
-
-
- Foundation for future governance systems
-
-
-
-
-
-
-
-
-
AI Governance Readiness Assessment
-
- Before implementing governance frameworks, organizations need honest answers to difficult questions.
- This assessment helps identify gaps, risks, and organizational readiness challenges.
+
+
+
+
+
+
The Governance Gap
+
+
+ Current AI governance approaches—policy documents, training programmes, ethical guidelines—rely on voluntary compliance. LLM systems can bypass these controls simply by not invoking them. When an AI agent needs to check a policy, it must choose to do so. When it should escalate a decision to human oversight, it must recognise that obligation.
+
+
+ This creates a structural problem: governance exists only insofar as the AI acknowledges it. For organisations subject to EU AI Act Article 14 (human oversight requirements) or deploying AI in high-stakes domains, this voluntary model is inadequate.
+
+
+ Tractatus explores whether governance can be made architecturally external—difficult to bypass not through better prompts, but through system design that places control points outside the AI's discretion.
+
-
-
+
+
+
Architectural Approach
-
-
-
-
-
-
-
-
-
-
Current AI Tool Inventory
-
Do you have clear visibility into what AI systems are already in use?
-
-
- □
- Have you catalogued all AI tools currently used across departments (ChatGPT, Claude, Copilot, internal LLMs, etc.)?
-
-
- □
- Do you know which employees are using AI tools for customer-facing communications, code generation, or decision support?
-
-
- □
- Can you identify which AI interactions involve proprietary data, customer information, or confidential business intelligence?
-
-
- □
- Do you have visibility into shadow AI usage (employees using personal accounts for work tasks)?
-
-
- □
- Have you documented which third-party vendors are using AI in services they provide to you?
-
-
-
-
+
+
+
+
Three-Layer Architecture
+
+
+
-
-
-
-
-
-
-
-
+
+
+
+ 1.
+
+ Agent Runtime Layer — Any LLM system (Claude Code, Copilot, custom agents, LangChain, CrewAI). The AI system being governed.
+
-
-
Strategic AI Deployment Plans
-
What are you planning to build, and have you assessed the governance implications?
-
-
- □
- Have you prioritized AI initiatives by risk level (customer-facing vs. internal, high-stakes vs. low-stakes)?
-
-
- □
- For each planned AI system, have you identified who is accountable when it makes a mistake or causes harm?
-
-
- □
- Do you have criteria for determining which decisions should remain human-controlled vs. AI-assisted vs. fully automated?
-
-
- □
- Have you evaluated whether your planned AI deployments fall under EU AI Act "high-risk" categories?
-
-
- □
- Can you articulate what "safe failure" looks like for each planned AI system?
-
-
+
+ 2.
+
+ Governance Layer — Six autonomous services that intercept, validate, and document AI operations. External to the AI runtime.
+
-
-
-
-
-
-
-
-
-
-
-
-
-
Workflow & Process Integration
-
How will AI fit into existing processes, and what breaks when it fails?
-
-
- □
- Have you mapped out which human roles will shift from "doer" to "reviewer/validator" of AI output?
-
-
- □
- Do you have processes to detect when employees are blindly accepting AI recommendations without validation?
-
-
- □
- Can your organization sustain critical operations if AI systems become unavailable for hours or days?
-
-
- □
- Have you considered the handoff points between AI-generated work and human-controlled processes (e.g., draft→review→approval)?
-
-
- □
- Do you know which workflows will require sequential AI operations, and how errors compound across multiple AI steps?
-
-
- □
- Have you assessed whether introducing AI will create new bottlenecks (e.g., senior staff spending all day reviewing AI output)?
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Decision Authority & Boundaries
-
Who decides what AI can and cannot do, and how are those boundaries enforced?
-
-
- □
- Have you defined which types of decisions AI systems are prohibited from making (even with human oversight)?
-
-
- □
- Do you have a governance board or designated owner responsible for AI safety and compliance decisions?
-
-
- □
- Can you enforce AI usage policies technically (not just via policy documents employees may ignore)?
-
-
- □
- Have you established clear escalation paths when AI systems encounter edge cases or ambiguous situations?
-
-
- □
- Do you have audit mechanisms to detect policy violations or unauthorized AI usage patterns?
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Incident Preparedness
-
What happens when AI systems fail, hallucinate, or cause harm?
-
-
- □
- Do you have incident response procedures specifically for AI failures (separate from general IT incidents)?
-
-
- □
- Can you trace AI-generated content or decisions back to specific prompts, model versions, and responsible parties?
-
-
- □
- Have you war-gamed scenarios where AI provides plausible-sounding but incorrect information that leads to business harm?
-
-
- □
- Do you have kill switches or rollback procedures to disable AI systems that are behaving unpredictably?
-
-
- □
- Have you assessed your liability exposure if AI systems discriminate, leak data, or violate regulations?
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Human & Cultural Readiness
-
Is your organization culturally prepared for the messy reality of AI governance?
-
-
- □
- Have you addressed employee fears about job displacement or skill obsolescence honestly?
-
-
- □
- Do your teams have the skills to critically evaluate AI output, or do they lack domain expertise to spot errors?
-
-
- □
- Are employees empowered to challenge or override AI recommendations without career risk?
-
-
- □
- Have you created incentives that reward thoughtful AI use over speed or cost savings alone?
-
-
- □
- Does your organization have realistic expectations about AI limitations, or is there pressure to treat it as infallible?
-
-
- □
- Have you allocated time and resources for governance work, or is it expected "on top of" existing responsibilities?
-
-
-
-
-
-
-
-
-
-
-
What Your Answers Reveal
-
-
- If you checked most boxes: You're ahead of most organizations but likely uncovering how complex AI governance truly is. The hard work is implementation and cultural change.
-
-
- If you checked some boxes: You have awareness but significant gaps. These gaps represent risk, but also clarity about where to focus governance efforts.
-
-
- If you checked few boxes: You're in good company—most organizations are here. The challenge is building governance capability while AI deployment accelerates around you.
-
-
- Note: This assessment is designed to provoke strategic thinking, not to sell you a solution. Effective AI governance requires organizational commitment, not just technology purchases. Tractatus is a research framework exploring architectural approaches to some of these challenges—it is not a comprehensive answer to all the questions above.
-
-
-
-
-
-
-
-
-
-
-
-
How Tractatus Works
-
- Six integrated components work together to provide structural AI governance
-
-
-
-
-
-
-
-
-
-
-
-
-
-
BoundaryEnforcer
-
Prevents AI systems from making values decisions without human approval. Ensures critical decisions remain under human control.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
InstructionPersistenceClassifier
-
Maintains organizational directives across AI interactions. Helps reduce instruction drift and inconsistency over time.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
CrossReferenceValidator
-
Validates AI decisions against established policies. Designed to detect potential conflicts before they occur.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
ContextPressureMonitor
-
Tracks AI session complexity and token usage. Helps prevent context degradation and maintains decision quality.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
MetacognitiveVerifier
-
Validates reasoning quality for complex operations. Designed to improve decision coherence and reduce errors.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
PluralisticDeliberationOrchestrator
-
Facilitates multi-stakeholder deliberation when values conflicts occur. Ensures non-hierarchical engagement and documents moral remainder.
-
-
-
-
-
-
-
-
-
-
-
Stakeholder Considerations
-
How different leadership roles may evaluate Tractatus
-
-
-
-
-
-
-
-
-
-
-
-
CEO / Founder
-
-
-
- •
- Structural approach to AI risk management
-
-
- •
- Potential competitive differentiation through governance
-
-
- •
- Framework for responsible AI deployment
-
-
-
-
-
-
-
-
-
-
-
-
-
CFO
-
-
-
- •
- Designed to help reduce regulatory fine risk (€35M max)
-
-
- •
- May reduce AI project failure costs (42% industry avg)
-
-
- •
- Research-stage framework (ROI not yet established)
-
-
-
-
-
-
-
-
-
-
-
-
-
CTO / Engineering
-
-
-
- •
- Architectural patterns for LLM safety controls
-
-
- •
- Reference implementation for governance agents
-
-
- •
- Adaptable to organizational tech stacks
-
-
-
-
-
-
-
-
-
-
-
-
-
CISO / Security
-
-
-
- •
- Controls for AI decision boundaries
-
-
- •
- Audit trails for AI actions
-
-
- •
- Risk mitigation through architectural controls
-
-
-
-
-
-
-
-
-
-
-
-
-
Legal / Compliance
-
-
-
- •
- Designed for EU AI Act alignment
-
-
- •
- Auditable decision-making framework
-
-
- •
- Documented governance processes
-
-
-
-
-
-
-
-
-
-
-
-
-
Product / AI Teams
-
-
-
- •
- Framework for safer AI feature deployment
-
-
- •
- Designed to reduce failure modes
-
-
- •
- Patterns for responsible AI innovation
-
-
-
-
-
-
-
-
-
-
-
Explore the Framework
-
Technical documentation and implementation resources