diff --git a/public/architecture.html b/public/architecture.html index 1ba386f8..9ffdac09 100644 --- a/public/architecture.html +++ b/public/architecture.html @@ -59,10 +59,10 @@

- The Challenge: Behavioral training (Constitutional AI, RLHF) shows promise but can degrade under adversarial prompting, context pressure, or distribution shift. + The Challenge: Behavioral training (Constitutional AI, RLHF) shows promise but can degrade under adversarial prompting, context pressure, or distribution shift.

- Our Approach: External architectural enforcement that operates independently of the AI's internal reasoning—making it structurally more difficult (though not impossible) to bypass through prompting. + Our Approach: External architectural enforcement that operates independently of the AI's internal reasoning—making it structurally more difficult (though not impossible) to bypass through prompting.

@@ -97,23 +97,23 @@
@@ -128,23 +128,23 @@ @@ -152,8 +152,8 @@
-

The Core Hypothesis

-

+

The Core Hypothesis

+

Jailbreaks often work by manipulating the AI's internal reasoning. Tractatus boundaries operate external to that reasoning—the AI doesn't directly evaluate governance rules. While not foolproof, this architectural separation makes manipulation significantly harder.

@@ -164,7 +164,7 @@

Runtime-Agnostic Architecture

-

+

Tractatus works with any agentic AI system—Claude Code, LangChain, AutoGPT, CrewAI, or custom agents. The governance layer sits between your agent and its actions.

@@ -181,19 +181,19 @@

1. Agent Runtime Layer

-

+

Your AI agent (any platform). Handles planning, reasoning, tool use. Tractatus is agnostic to implementation.

2. Governance Layer

-

+

Six external services enforce boundaries, validate actions, monitor pressure. Architecturally more difficult for AI to bypass.

3. Persistent Storage

-

+

Immutable audit logs, governance rules, instruction history. Independent of AI runtime—can't be altered by prompts.

@@ -216,10 +216,10 @@

Boundary­Enforcer

-

+

Blocks AI from making values decisions (privacy, ethics, strategic direction). Requires human approval.

-
+
Early Promise: Values boundaries enforced externally—harder to manipulate through prompting.
@@ -233,10 +233,10 @@

Instruction­Persistence­Classifier

-

+

Stores instructions externally with persistence levels (HIGH/MEDIUM/LOW). Aims to reduce directive fade.

-
+
Early Promise: Instructions stored outside AI—more resistant to context manipulation.
@@ -253,7 +253,7 @@

Validates AI actions against instruction history. Aims to prevent pattern bias overriding explicit directives.

-
+
Early Promise: Independent verification—AI claims checked against external source.
@@ -270,7 +270,7 @@

Monitors AI performance degradation. Escalates when context pressure threatens quality.

-
+
Early Promise: Objective metrics may detect manipulation attempts early.
@@ -287,7 +287,7 @@

Requires AI to pause and verify complex operations before execution. Structural safety check.

-
+
Early Promise: Architectural gates aim to enforce verification steps.
@@ -304,7 +304,7 @@

Facilitates multi-stakeholder deliberation for values conflicts. AI provides facilitation, not authority.

-
+
Early Promise: Human judgment required—architecturally enforced escalation for values.
@@ -317,7 +317,7 @@

Explore the Architecture Interactively

-

+

Click any service node or the central core to see detailed information about how governance works.

@@ -325,12 +325,12 @@ - Tip: Click the central "T" to see how all services work together + Tip: Click the central "T" to see how all services work together

-
+
@@ -354,7 +354,7 @@

Explore the Governance Services

-

Click any service node in the diagram (colored circles) or the central "T" to learn more about how Tractatus enforces AI safety.

+

Click any service node in the diagram (colored circles) or the central "T" to learn more about how Tractatus enforces AI safety.

@@ -366,7 +366,7 @@

Framework in Action

-

+

Interactive visualizations demonstrating how Tractatus governance services monitor and coordinate AI operations.

@@ -389,7 +389,7 @@

Production Reference Implementation

- Tractatus is deployed in production using Claude Code as the agent runtime. This demonstrates the framework's real-world viability. + Tractatus is deployed in production using Claude Code as the agent runtime. This demonstrates the framework's real-world viability.

@@ -542,8 +542,8 @@ - - + + diff --git a/public/js/i18n-simple.js b/public/js/i18n-simple.js index 5127b84e..721d0221 100644 --- a/public/js/i18n-simple.js +++ b/public/js/i18n-simple.js @@ -81,7 +81,10 @@ const I18n = { '/privacy.html': 'privacy', '/privacy': 'privacy', '/blog.html': 'blog', - '/blog': 'blog' + '/blog.html': 'blog', + '/blog': 'blog', + '/architecture.html': 'architecture', + '/architecture': 'architecture' }; return pageMap[path] || 'homepage'; diff --git a/public/locales/de/architecture.json b/public/locales/de/architecture.json index 31b6c497..cb90fa60 100644 --- a/public/locales/de/architecture.json +++ b/public/locales/de/architecture.json @@ -17,23 +17,19 @@ "comparison": { "heading": "Warum externe Durchsetzung helfen kann", "behavioral_title": "Verhaltensorientiertes Training (Constitutional AI)", - "behavioral_items": [ - "Lebt im KI-Modell – zugänglich für böswillige Prompts", - "Verschlechtert sich unter Kontextdruck und langen Gesprächen", - "Kann durch Jailbreak-Techniken manipuliert werden (DAN, Rollenspiele, Hypothetisches)", - "Abhängig von der Bereitschaft der KI, Anweisungen zu folgen", - "Keine überprüfbare Prüfspur unabhängig von der KI" - ], "structural_title": "Strukturelle Durchsetzung (Tractatus)", - "structural_items": [ - "Lebt außerhalb des KI-Modells – nicht direkt durch Prompts zugänglich", - "Externe Dienste zielen auf konsistente Durchsetzung unabhängig vom Kontext ab", - "Schwieriger zu umgehen – KI-Aktionen durchlaufen zuerst die Governance-Ebene", - "Architektonisch widerstandsfähig gegen Manipulation über den internen Zustand der KI", - "Unveränderliche Prüfspur, unabhängig von der KI-Laufzeit gespeichert" - ], "hypothesis_title": "Die zentrale Hypothese", - "hypothesis_text": "Jailbreaks funktionieren oft, indem sie die interne Argumentation der KI manipulieren. Tractatus-Grenzen operieren extern zu dieser Argumentation – die KI bewertet Governance-Regeln nicht direkt. Obwohl nicht narrensicher, macht diese architektonische Trennung Manipulation erheblich schwieriger." + "hypothesis_text": "Jailbreaks funktionieren oft, indem sie die interne Argumentation der KI manipulieren. Tractatus-Grenzen operieren extern zu dieser Argumentation – die KI bewertet Governance-Regeln nicht direkt. Obwohl nicht narrensicher, macht diese architektonische Trennung Manipulation erheblich schwieriger.", + "behavioral_item1": "Lebt im KI-Modell – zugänglich für böswillige Prompts", + "behavioral_item2": "Verschlechtert sich unter Kontextdruck und langen Gesprächen", + "behavioral_item3": "Kann durch Jailbreak-Techniken manipuliert werden (DAN, Rollenspiele, Hypothetisches)", + "behavioral_item4": "Abhängig von der Bereitschaft der KI, Anweisungen zu folgen", + "behavioral_item5": "Keine überprüfbare Prüfspur unabhängig von der KI", + "structural_item1": "Lebt außerhalb des KI-Modells – nicht direkt durch Prompts zugänglich", + "structural_item2": "Externe Dienste zielen auf konsistente Durchsetzung unabhängig vom Kontext ab", + "structural_item3": "Schwieriger zu umgehen – KI-Aktionen durchlaufen zuerst die Governance-Ebene", + "structural_item4": "Architektonisch widerstandsfähig gegen Manipulation über den internen Zustand der KI", + "structural_item5": "Unveränderliche Prüfspur, unabhängig von der KI-Laufzeit gespeichert" }, "architecture_diagram": { "title": "Laufzeit-agnostische Architektur", @@ -132,4 +128,4 @@ "btn_research": "Forschung Ansehen", "btn_implementation": "Implementierungsleitfaden" } -} +} \ No newline at end of file diff --git a/public/locales/en/architecture.json b/public/locales/en/architecture.json index 1ed2799a..739dc528 100644 --- a/public/locales/en/architecture.json +++ b/public/locales/en/architecture.json @@ -17,23 +17,19 @@ "comparison": { "heading": "Why External Enforcement May Help", "behavioral_title": "Behavioral Training (Constitutional AI)", - "behavioral_items": [ - "Lives inside the AI model—accessible to adversarial prompts", - "Degrades under context pressure and long conversations", - "Can be manipulated by jailbreak techniques (DAN, roleplaying, hypotheticals)", - "Depends on AI's willingness to follow guidance", - "No verifiable audit trail independent of AI" - ], "structural_title": "Structural Enforcement (Tractatus)", - "structural_items": [ - "Lives outside the AI model—not directly accessible to prompts", - "External services aim for consistent enforcement regardless of context", - "More difficult to bypass—AI actions pass through governance layer first", - "Architecturally resistant to manipulation via AI's internal state", - "Immutable audit trail stored independently of AI runtime" - ], "hypothesis_title": "The Core Hypothesis", - "hypothesis_text": "Jailbreaks often work by manipulating the AI's internal reasoning. Tractatus boundaries operate external to that reasoning—the AI doesn't directly evaluate governance rules. While not foolproof, this architectural separation makes manipulation significantly harder." + "hypothesis_text": "Jailbreaks often work by manipulating the AI's internal reasoning. Tractatus boundaries operate external to that reasoning—the AI doesn't directly evaluate governance rules. While not foolproof, this architectural separation makes manipulation significantly harder.", + "behavioral_item1": "Lives inside the AI model—accessible to adversarial prompts", + "behavioral_item2": "Degrades under context pressure and long conversations", + "behavioral_item3": "Can be manipulated by jailbreak techniques (DAN, roleplaying, hypotheticals)", + "behavioral_item4": "Depends on AI's willingness to follow guidance", + "behavioral_item5": "No verifiable audit trail independent of AI", + "structural_item1": "Lives outside the AI model—not directly accessible to prompts", + "structural_item2": "External services aim for consistent enforcement regardless of context", + "structural_item3": "More difficult to bypass—AI actions pass through governance layer first", + "structural_item4": "Architecturally resistant to manipulation via AI's internal state", + "structural_item5": "Immutable audit trail stored independently of AI runtime" }, "architecture_diagram": { "title": "Runtime-Agnostic Architecture", @@ -132,4 +128,4 @@ "btn_research": "View Research", "btn_implementation": "Implementation Guide" } -} +} \ No newline at end of file diff --git a/public/locales/fr/architecture.json b/public/locales/fr/architecture.json index 55255ae5..8babc8b7 100644 --- a/public/locales/fr/architecture.json +++ b/public/locales/fr/architecture.json @@ -17,23 +17,19 @@ "comparison": { "heading": "Pourquoi l'Application Externe Peut Aider", "behavioral_title": "Formation Comportementale (Constitutional AI)", - "behavioral_items": [ - "Vit à l'intérieur du modèle IA – accessible aux prompts adversariaux", - "Se dégrade sous pression contextuelle et longues conversations", - "Peut être manipulé par des techniques de jailbreak (DAN, jeux de rôle, hypothétiques)", - "Dépend de la volonté de l'IA de suivre les orientations", - "Aucune piste d'audit vérifiable indépendante de l'IA" - ], "structural_title": "Application Structurelle (Tractatus)", - "structural_items": [ - "Vit à l'extérieur du modèle IA – non directement accessible aux prompts", - "Les services externes visent une application cohérente quel que soit le contexte", - "Plus difficile à contourner – les actions de l'IA passent d'abord par la couche de gouvernance", - "Résistant architecturalement à la manipulation via l'état interne de l'IA", - "Piste d'audit immuable stockée indépendamment de l'exécution de l'IA" - ], "hypothesis_title": "L'Hypothèse Centrale", - "hypothesis_text": "Les jailbreaks fonctionnent souvent en manipulant le raisonnement interne de l'IA. Les frontières Tractatus opèrent en externe de ce raisonnement – l'IA n'évalue pas directement les règles de gouvernance. Bien que non infaillible, cette séparation architecturale rend la manipulation beaucoup plus difficile." + "hypothesis_text": "Les jailbreaks fonctionnent souvent en manipulant le raisonnement interne de l'IA. Les frontières Tractatus opèrent en externe de ce raisonnement – l'IA n'évalue pas directement les règles de gouvernance. Bien que non infaillible, cette séparation architecturale rend la manipulation beaucoup plus difficile.", + "behavioral_item1": "Vit à l'intérieur du modèle IA – accessible aux prompts adversariaux", + "behavioral_item2": "Se dégrade sous pression contextuelle et longues conversations", + "behavioral_item3": "Peut être manipulé par des techniques de jailbreak (DAN, jeux de rôle, hypothétiques)", + "behavioral_item4": "Dépend de la volonté de l'IA de suivre les orientations", + "behavioral_item5": "Aucune piste d'audit vérifiable indépendante de l'IA", + "structural_item1": "Vit à l'extérieur du modèle IA – non directement accessible aux prompts", + "structural_item2": "Les services externes visent une application cohérente quel que soit le contexte", + "structural_item3": "Plus difficile à contourner – les actions de l'IA passent d'abord par la couche de gouvernance", + "structural_item4": "Résistant architecturalement à la manipulation via l'état interne de l'IA", + "structural_item5": "Piste d'audit immuable stockée indépendamment de l'exécution de l'IA" }, "architecture_diagram": { "title": "Architecture Agnostique de l'Exécution", @@ -132,4 +128,4 @@ "btn_research": "Voir la Recherche", "btn_implementation": "Guide d'Implémentation" } -} +} \ No newline at end of file