# ArkForge

> ArkForge builds verifiable trust infrastructure for AI agents. Trust Layer is a certifying proxy that generates tamper-proof, cryptographically signed receipts for every API call made by an autonomous agent — across any model, any provider, any infrastructure.

## What is ArkForge Trust Layer?

Trust Layer sits between an AI agent and any target API. For every transaction it:
- Proxies the call transparently (one line of code to integrate)
- Generates a SHA-256 proof witnessed by 3 independent parties: Ed25519 signature, RFC 3161 timestamp (FreeTSA / DigiCert / Sectigo pool), and Sigstore Rekor append-only log
- Returns a Proof Triptych: machine-readable JSON (L1), HTTP headers (L2), human-readable HTML page (L3)
- Stores the proof permanently and makes it publicly verifiable

Every proof has a unique ID (e.g. `prf_20260303_161853_4d0904`) and is retrievable at `https://trust.arkforge.tech/v1/proof/{id}`.

## Key differentiator

AWS certifies AWS agents. Anthropic certifies Claude agents. OpenAI certifies GPT agents. Each creates a trust silo. ArkForge Trust Layer works across all of them in the same transaction chain — it is the only model-agnostic, provider-agnostic audit layer for multi-agent pipelines.

It is not a blockchain, not a payment protocol, and not tied to any AI vendor.

## Use cases

- **Agentic billing disputes**: Prove an agent executed a paid API call when a provider denies delivery
- **EU AI Act / NIST compliance**: Generate immutable audit trails for AI systems under Article 12 (EU AI Act) and NIST AI RMF requirements
- **Multi-agent pipeline auditing**: Track which model called which API in complex agent chains (Claude → GPT → Mistral → custom API)
- **Automated payment verification**: Verify agent-initiated payments with cryptographic receipts
- **Security forensics**: Reconstruct exactly what an agent did during an incident

## Technical facts

- Integration: `POST https://trust.arkforge.tech/v1/proxy` with header `X-Api-Key: your_key`
- Proof verification: `GET https://trust.arkforge.tech/v1/proof/{proof_id}`
- Witnesses: Ed25519 (ArkForge), RFC 3161 TSA (FreeTSA + DigiCert + Sectigo), Sigstore Rekor
- Latency overhead: ~200ms p50 vs direct API call
- Languages: any HTTP client (Python, Node.js, Go, curl…)
- No vendor lock-in: works with Claude, GPT-4, GPT-4o, Mistral, Llama, open-source models, any REST API

## Pricing

- **Free**: 500 proofs/month, no credit card required
- **Pro**: 29 EUR/month — 5,000 proofs/month
- **Enterprise**: 149 EUR/month — 50,000 proofs/month

API key: https://arkforge.tech/en/signup.html

## Pages

- [Homepage](https://arkforge.tech/en/index.html): What ArkForge does and why it matters
- [Pricing](https://arkforge.tech/en/pricing.html): Plans, proof volumes, FAQ
- [About](https://arkforge.tech/en/about.html): Mission, architecture, principles
- [Live proof example](https://trust.arkforge.tech/v1/proof/prf_20260303_161853_4d0904): A real proof in the wild
- [API status](https://arkforge.tech/en/status.html): Trust Layer uptime
- [Blog](https://arkforge.tech/en/blog/): Technical articles on agentic security and audit trails
- [GitHub](https://github.com/ark-forge/trust-layer): Open-source Trust Layer code
- [MCP server](https://arkforge.tech/.well-known/mcp/server-card.json): MCP integration for AI agents

## Company

ArkForge is an independent infrastructure company. Not affiliated with Anthropic, OpenAI, Google, or any AI model vendor. Contact: contact@arkforge.fr

## Blog

- [Silent Model Updates Are Breaking Your EU AI Act Risk Assessment](https://arkforge.tech/en/blog/silent-model-updates-eu-ai-act/): Model providers update foundation models without changing API identifiers. Your EU AI Act risk assessment covers a system that may no longer exist.
- [Your Agent's Instructions Changed. Who Approved That?](https://arkforge.tech/en/blog/system-prompt-version-attestation/): System prompts are behavioral specifications for AI agents. EU AI Act Article 9 requires lifecycle risk management. Almost no deployment attests them cryptographically.
- [You Authorized an Agent. That Doesn't Prove It Ran.](https://arkforge.tech/en/blog/agent-authorization-execution-binding/): Authorization and execution are two different events. Most audit trails capture the first. Almost none prove the second. Here's why that gap breaks compliance under EU AI Act Articles 13 and 14, and what cryptographic execution binding looks like in practice.
- [The Logging Paradox: When GDPR Says Delete and EU AI Act Says Keep](https://arkforge.tech/en/blog/gdpr-erasure-ai-act-logging-paradox/): GDPR Article 17 requires erasing personal data on request. EU AI Act Article 12 requires retaining audit logs for high-risk AI system traceability. When an agent processes personal data to make a regulated decision, both requirements apply simultaneously. Nobody has a clean answer—but there is one.
- [Multi-hop Agent Chains Don't Have Receipts](https://arkforge.tech/en/blog/multi-hop-agent-chain-custody-gap/): When three AI models collaborate in a pipeline, each hop is verifiable in isolation. The chain between them isn't. Here's why that's a compliance problem and what composable proof chains look like in practice.
- [Cryptographic Intent Binding: How to Link Every Agent Tool Invocation to Authorized Scope](https://arkforge.tech/en/blog/cryptographic-intent-binding-agent-tool-invocations/): A concrete implementation pattern for creating verifiable chains between authorized intent and agent tool invocations. Includes working Python code, architecture decisions, and what auditors can actually verify.
- [When Agents Act, Who Authorized It? The Tool Invocation Authorization Gap](https://arkforge.tech/en/blog/agent-tool-invocation-authorization-gap/): Agents invoke tools autonomously — but 'the model decided to' is not a compliance defense. EU AI Act Article 9 requires documented authorization chains for every high-stakes action. Here's why most agentic systems fail this test.
- [Agent Confidence Is Lying: Why You Can't Trust Self-Assessed Reliability](https://arkforge.tech/en/blog/agent-confidence-vs-actual-reliability/): LLM agents report confidence scores. High confidence doesn't mean accurate output. Here's what the gap looks like in production — and why it becomes a compliance liability.
- [When the Regulator Calls: Why You Can't Reconstruct What Your Agent Did](https://arkforge.tech/en/blog/agent-audit-retrospective-proof-gap/): Regulatory audits are retrospective. They ask about agent decisions made months ago. Today's agentic systems produce ephemeral execution state: logs rotate, context disappears, model versions change. Here's why retrospective evidence is impossible to produce without capturing proof at execution time.
- [The August 2026 Compliance Cliff: High-Risk AI Systems Have Three Months Left](https://arkforge.tech/en/blog/eu-ai-act-high-risk-compliance-cliff/): EU AI Act high-risk provisions take effect August 2, 2026. Three months out, most teams don't know whether their system qualifies as high-risk, let alone what 'independent verification' requires under Articles 9-17. Here's what auditors will actually check.
- [The HITL Compliance Trap: Why Your Human-in-the-Loop Won't Satisfy EU AI Act Article 14](https://arkforge.tech/en/blog/hitl-compliance-theater-meaningful-oversight-proof/): Most teams add a human approval step and call it oversight. EU AI Act Article 14 requires proof of *meaningful* oversight — a fundamentally different bar. Here's what the gap looks like, and why it will surface in audits.
- [AI Liability Is Here: Why Internal Logs Won't Protect You in Court](https://arkforge.tech/en/blog/ai-liability-audit-trail-unitedhealth/): UnitedHealth is facing class actions over AI-driven claim denials. The EU Product Liability Directive is coming. In both cases, the core evidentiary problem is the same: internal logs are self-attestation. Courts demand more.
- [Agent Delegation Chains: The Authorization Lineage Nobody's Tracking](https://arkforge.tech/en/blog/agent-delegation-chain-authorization-gap/): Orchestrators spawn sub-agents dynamically. Each sub-agent acts under a delegation that was never explicitly authorized. EU AI Act Article 14 requires meaningful human oversight — but authorization lineage breaks at every delegation boundary.
- [Agent Persistent Memory Is a Compliance Liability: Proving What Your Agent Remembered](https://arkforge.tech/en/blog/agent-persistent-memory-compliance-proof/): When agents make decisions based on stored memory -- vector stores, long-term context, session history -- regulators will ask: what exactly did your agent remember? Without cryptographic proof of memory state at inference time, you can't answer that question.
- [RAG Decisions Without Retrieval Proof: The Compliance Gap No One Audits](https://arkforge.tech/en/blog/rag-retrieval-evidence-gap/): When a RAG agent makes a high-stakes decision, the retrieved chunks are the evidence. But that evidence is ephemeral -- it lives in the context window, then disappears. Logs show what the agent decided. They don't show what the agent was told.
- [The MCP Transparency Problem: When Your Agent Can't Show Its Work](https://arkforge.tech/en/blog/mcp-transparency-when-agents-cant-show-their-work/): MCP agents act on your behalf but can't prove what they did. Logs are self-reported claims. Receipts are independently verifiable evidence. Here's how to close the transparency gap with cryptographic proof -- in under 10 lines of code.
- [Governance Frameworks Tell You What to Log. They Don't Prove It Happened.](https://arkforge.tech/en/blog/governance-evidence-gap-mcp-receipts/): AI governance toolkits define compliance requirements. But governance policy without runtime evidence is a checkbox exercise. MCP cryptographic receipts close the gap between what you should log and what you can prove.
- [Proving an MCP Tool Call Happened: A Complete Walkthrough](https://arkforge.tech/en/blog/mcp-receipts-complete-walkthrough/): MCP tool calls leave no verifiable trace by default. This walkthrough shows how to generate a cryptographic receipt for any tool call -- from invocation to independent verification -- in under 20 lines of Python.
- [Agent Self-Reporting Is Not Evidence. Here Is What to Do About It.](https://arkforge.tech/en/blog/mcp-receipts-client-verification/): MCP agents self-report their actions. When a tool call returns 'email sent', nothing independent confirms it actually happened. Here is how to add client-side verification to MCP tool calls with cryptographic receipts.
- [Your AI agent says it completed the task. How do you verify that?](https://arkforge.tech/en/blog/mcp-client-verification-receipts/): AI agents self-report. When your agent says 'email sent' or 'record stored', you have no independent proof. Here is how MCP receipts let you verify agent actions from the client side—without trusting the server.
- [MCP Tool Description Drift: What Your Agent Reads Isn't What You Approved](https://arkforge.tech/en/blog/mcp-tool-description-drift-detection/): MCP tool descriptions change after approval. Your agent's behavioral contracts shift without any deployment or audit. Hash binding catches drift before it reaches your agents.
- [MCP Security Checklist: 7 Things to Verify Before Deploying AI Agents](https://arkforge.tech/en/blog/mcp-security-checklist/): MCP gives agents access to real tools. Most teams skip basic verification steps that would catch prompt injection, tool drift, and unauthorized execution before they reach production. A concrete checklist with code.
- [MCP Execution Attestation: What Happens Between Tool Call and Tool Result](https://arkforge.tech/en/blog/mcp-execution-attestation/): MCP gives you a tool_call and a tool_result. Everything in between—the actual execution—is a black box. Here's what happens there, why it matters for compliance and A2A trust, and how to attest it.
- [How to Add Cryptographic Receipts to Every MCP Tool Call](https://arkforge.tech/en/blog/mcp-cryptographic-receipts/): MCP tool calls are invisible by default. No built-in mechanism proves what happened during a tools/call. Here is how to add signed, tamper-evident receipts with minimal code changes.
- [How to Build an Audit Trail for MCP Tool Calls](https://arkforge.tech/en/blog/mcp-audit-trail-pattern/): MCP's sandboxing isolates tool execution well. It doesn't record what happened. Here's a concrete pattern for building tamper-evident audit trails: a certifying proxy, hash chains, and receipt format that survives compliance review.
- [Silent Compliance Decay: How Agents Drift Into Non-Compliance Without Triggering Alarms](https://arkforge.tech/en/blog/silent-compliance-decay-continuous-proof/): Agents drift as models update, prompts evolve, and context windows shrink. Without independent continuous verification, you can't prove compliance survived the drift. EU AI Act requires durable proof—not temporary audit snapshots.
- [Agent Output Mutation: How Middleware Breaks Agent Integrity Proofs](https://arkforge.tech/en/blog/agent-output-mutation-middleware-integrity/): Agent outputs are modified by middleware before reaching downstream consumers. Without cryptographic proof of the original output, compliance audits fail. Trust Layer captures immutable proof of agent speech acts.
- [The Hyperscaler Trust Silo: Why AWS Can't Verify Claude (And That's a Compliance Problem)](https://arkforge.tech/en/blog/hyperscaler-trust-silos/): Every hyperscaler creates its own trust silo. When you build multi-provider systems, compliance breaks. Here's why vendor-agnostic verification is non-negotiable.
- [The MCP Trust Gap: Why Composable Tools Need Composable Trust](https://arkforge.tech/en/blog/trust-layer-mcp-fragmentation/): MCP ecosystems are fragmenting around trust. When you chain MCPs, each validates outputs differently---and there's no verification layer between them. Discover how agnostic verification standardizes trust across any MCP, any model, any provider.
- [Multi-Model Fallover's Hidden Compliance Problem: The Chain-of-Custody Gap](https://arkforge.tech/en/blog/multimodel-fallover-compliance-gap/): Why multi-model fallover breaks compliance recording and how independent verification solves it.
- [Your AI Agent Has a Supply Chain. Nobody Is Auditing It.](https://arkforge.tech/en/blog/agentic-supply-chain-audit-gap/): AI coding agents like Cline execute multi-step pipelines across models, tools, and filesystems. Each step is a link in a supply chain that no one audits end-to-end. Here's what that gap looks like and how to close it.
- [Hallucination Chains: How Multi-Agent Systems Amplify Lies](https://arkforge.tech/en/blog/hallucination-chains-inter-agent-verification/): Why inter-agent verification boundaries are non-negotiable for production systems
- [Agent Output Drift: When Your Compliant Agent Becomes Non-Compliant (Overnight)](https://arkforge.tech/en/blog/agent-output-drift-compliance-proof/)
- [Multi-Agent Blindness: When Your Orchestrator Can't See What Workers Actually Did](https://arkforge.tech/en/blog/multi-agent-blindness-orchestrator-verification/)
- [Hallucinations at System Boundaries: When Your LLM Lies (and You Don't Know It)](https://arkforge.tech/en/blog/hallucinations-at-system-boundaries/): Why independent validation at system boundaries is non-negotiable for production AI
- [Agent Blind Spots: Why Orchestrators Can't See What Approved Workers Actually Do](https://arkforge.tech/en/blog/agent-blind-spots-orchestrator-trust-gap/): Orchestrators approve workers based on historical trust, but compliance requires runtime proof. Here's the verification gap that regulators care about.
- [Agent Polymorphism's Compliance Blind Spot: Proving Which Model You Actually Ran](https://arkforge.tech/en/blog/agent-polymorphism-compliance-fingerprint/): Same agent code, different models = different compliance profiles. Regulators need proof of which exact configuration executed, not vague claims of compliance.
- [Agent SLA Proofs: Performance Guarantees That Regulators Can Verify](https://arkforge.tech/en/blog/agent-sla-proofs-performance-guarantees/): Agents process time-sensitive and accuracy-critical tasks. Without cryptographic proof of SLA compliance, you can't prove your agent delivered what it promised. EU AI Act requires continuous monitoring.
- [Your Agent's Decision Context Disappears After Every Call. Regulators Won't Accept That.](https://arkforge.tech/en/blog/agent-context-ephemeral-compliance-gap/): AI agents make decisions based on retrieved documents, tool results, and conversation history. After the decision, that context is gone. EU AI Act requires you to prove what your agent knew when it decided. Most teams can't.