Governed trust infrastructure for AI agents — certification, governance, networking, and memory in one system.
How it works, simply: An AI agent is created. It needs to pass TBN certification — 6 security checks — and get a signed identity before it can do anything. Then TBN Governance is applied (rules, monitoring, compliance). Then the agent enters the AI Net — a highway where all governed agents live and operate, with 22 enforcement engines watching every action. Agent Memory supports the agent throughout — signed, poison-proof knowledge it can trust. The result: every agent is certified, governed, networked, and remembers honestly.
The Four Pillars
🔏
TBN Protocol
Agent registration + 6 challenge gates (Prompt Injection, Data Boundary, Hallucination, Budget/Fees, Sensitive Data, Instruction-following) + cryptographic signing, receipts, verification. The entry point — nothing acts without passing TBN first.
⚖️
TBN Governance
Trust Ledger (immutable action log), Rules Engine (what agents can/cannot do), Behaviour Monitor (real-time anomaly detection), and Compliance Reports (auto-generated GDPR + EU AI Act). The governance brain.
🌐
AI Net
22-engine enforcement pipeline (7 always-on + 15 conditional). Where registered agents live and operate. Every action evaluated. Returns PERMIT / HOLD / BLOCK.
🧠
Agent Memory
Hardin Memory — signed, poison-proof, temporally versioned, per-agent isolated. The only agent memory that's provably untampered.
How an AI action flows through TBN
1
TBN Protocol — Certification & Identity
An agent is created. It must pass 6 security challenges (Prompt Injection, Data Boundary, Hallucination, Budget/Fees, Sensitive Data, Instruction-following), get certified, and receive a signed identity. Nothing operates without this.
▼
2
TBN Governance — Rules & monitoring applied
Governance is applied to the certified agent: Trust Ledger records every action immutably. Rules Engine defines what's allowed. Behaviour Monitor watches for anomalies. Compliance Reports auto-generate for regulators.
▼
3
AI Net — The highway where agents live
The agent enters the AI Agent Network — a highway of governed agents. 22 enforcement engines (7 always-on + 15 conditional) evaluate every action in real time. PERMIT / HOLD / BLOCK. The agent stays here, operating under constant oversight.
↕ supports throughout▼
4
Agent Memory — Signed knowledge that supports the agent
Throughout all of this, the agent's memory is always available — signed, poison-proof, temporally versioned, isolated per agent. The agent recalls only verified facts. Corrupted or forged knowledge is silently rejected.
▼
5
Result — Signed receipt + Governed Action Record
Every action produces a signed, tamper-evident, independently-verifiable record. Nine governance layers bound in one record. Anyone can check it at a public URL. No trust in us required.
Agent Memory — Hardin Memory Engine (v2)
Every fact an agent knows is cryptographically signed, hash-chained, and independently verifiable. The only AI memory where you can prove what the agent believed, when it changed its mind, and that nobody altered the record.
🔏 Signed & ChainedEvery memory shard is RSA-PSS-SHA256 signed. Recall re-verifies before trusting. Forged memories are silently rejected.
🛡️ Poison-ProofContradiction detection gate on write. Nobody can teach the agent a lie by repetition.
⏱️ Temporal VersioningNEWFacts update honestly — new signed fact supersedes old. Nothing erased. Full version timeline provable on demand.
🔒 Per-Agent IsolationNEWEach agent gets its own sealed memory file. One agent's knowledge can never leak into or be poisoned by another's. Structural, not a filter.
🏷️ Entity-Aware RecallNEW"What do we know about X?" as a direct lookup, not a semantic guess. Named entities auto-linked on every write.
🧹 Near-Duplicate DedupNEWSame fact reworded? Detected and deduplicated at write time. No storage bloat, no false version chains.
🧠 3-Tier TrustFacts (signed, immutable) / Preferences (mutable, low-trust) / Learned Policy (gated, human-approved). Adaptivity can never grant a right.
✓ Whole-Store AuditOne call: Merkle root over all shards, chain integrity check, signature verification on every entry. "Is the entire memory intact?" — provable.
TBN Governance — 4 Features
The governance brain. Logs everything, enforces rules, detects rogue behaviour, and auto-generates compliance evidence for regulators.
1. Trust Ledger
Immutable, hash-chained record of every agent action — like a black box. Tampering with any entry breaks the chain and is instantly detectable.
2. Rules Engine
Define what agents can and cannot do. Block data exfiltration, rate-limit calls, restrict after-hours access, require certification for sensitive data. BLOCK / THROTTLE / ALERT.
3. Behaviour Monitor
Real-time anomaly detection. High call rates, failure spikes, multiple IPs (possible cloning), off-hours access — flagged automatically with severity levels.
4. Compliance Reports
Auto-generated GDPR and EU AI Act compliance reports from real ledger data. Scored per-article, signed, with evidence from the Trust Ledger.
AI Net — 22 Engines, 2 Wheels
The enforcement layer. Where agents live after TBN registration. Every action is evaluated before it can proceed. Wheel 1 always runs. Wheel 2 fires conditionally based on risk/context. The agent cannot bypass the network — it's architectural, not optional.
The Governed Action Record — 9 layers, one signed record
1 Identity who acted
2 Mandate under whose authority
3 Decision-state what it knew
4 Governance permit / hold / block
5 Memory knowledge used
6 Outcome what it did
7 Responsibility who is liable
8 Evidence hash + signature
9 Chain + scope tamper-evident + honest limits
Why this wins
Memory platforms (Mem0, Zep, Supermemory) store and retrieve but don't sign, don't version, and can't prove integrity. Agent frameworks (LangChain, CrewAI) orchestrate but don't govern. Observability tools (LangSmith) log but can't enforce. Compliance tools (C2PA) mark content but don't govern what produced it.
TBN is the only platform where an AI agent's identity, memory, decisions, and inter-agent communication are all cryptographically governed and independently verifiable — in one system. That is the difference between "we monitor our AI" and "we govern our AI, and here is the proof."
Live now
TBN Protocol: tbn.hardinai.co.uk Agent Memory: memory.hardinai.co.uk AI Net: net.hardinai.co.uk Governance Tracer: tbn.hardinai.co.uk/tracer Governance Dashboard: tbn.hardinai.co.uk/governance Verify any action: tbn.hardinai.co.uk/gar/<record_id>