Secure LLMs · the sovereign tier

We don’t ask you to trust the model.

We run open-weight models inside your perimeter with no route to the internet and no name to resolve. The weights are proven byte-identical to the public artifact. Every call is anchored to a tamper-evident chain. You get the evidence, not our word for it.

The substrate is why you’re allowed to buy the application.
See it carrying a real one: Hyfstele MLR for pharma promotional review →

What we do

We prove it’s byte-identical to the public artifact, so no backdoor can be aimed at you. We enumerate its dormant capacity and prune what never fires. We give it no channel to talk. And we record everything it does.

Four independent controls. Each one is verifiable on your own hardware, by your own people, without taking our word for anything — or the model publisher’s.

Four independent controls

Each one verifiable. None of them asks for trust.

01
Attest

Hash every weight file against the digest the publisher posted publicly. Anything surviving byte-identity is in the artifact the whole world runs — so it can’t be aimed at you.

Verified on real bytes
02
Enumerate

A backdoor hides in capacity ordinary traffic never touches. Routing is discrete and countable, so we histogram it and ablate what never fires. Deleting the hiding place beats auditing it.

Built · not yet run on real weights
03
Contain

The enclave initiates zero outbound connections and resolves zero names. Not an allow-list. Not a policy. The capability does not exist.

Proven · survives reboot
04
Record

Every input, output, and routing fingerprint anchored to a tamper-evident chain. When prevention fails, detection survives — and corrections supersede in the record rather than replacing it.

Live
01 — Attestation

Sameness, not safety.

Proving a model is clean is impossible. Proving it is the same bytes everyone else is running is a hash comparison. That substitution is the most valuable move in this architecture, and it is widely missed.

We hash every weight file of our own copy against the digest the publisher has posted publicly, built as a Merkle tree at file granularity — so a mismatch localises to a specific tensor file rather than condemning the artifact wholesale.

Verified. Our archived copy’s tokenizer.json hashes to 0614fe83cadab421…, matching the published digest exactly. Full-artifact verification across all 22 weight shards is a mechanical extension of the same check.

What this does not prove. Sameness is not safety. It says nothing about whether the public artifact is itself compromised. It also treats the publisher’s posted digests as the reference — which is why the revision must be pinned and re-verified over time, so upstream mutation is detected rather than inherited.

Why this is decisive:

  • Any backdoor that survives byte-identity is present in the public artifact — the one downloaded by every lab, government, and red team on earth.
  • It therefore cannot be targeted at a specific customer. A targeted supply-chain substitution — the realistic threat — is eliminated, not mitigated.
  • The residual collapses from “unknowable property of 120B floats” into “shared fate with the open-source ecosystem” — a risk every organisation already accepts implicitly.
We’re not asking you to trust the publisher. We’re proving you run the identical artifact everyone else is auditing.
03 — Containment

Things go in. Nothing comes out.

A backdoor is only worth building if it has a channel: data out, or a trigger in. The Blackhole removes the channel.

The naive version is an allow-list — “the model may only talk to your internal systems.” That is not a guarantee. Those systems reach the internet, so a trickle through them reaches it too. Transitive egress is egress. We don’t police the capability. We remove it.

no external IP  +  no NAT  →  the 0.0.0.0/0 route is inert
deny-all egress firewall, logged
no name resolution
stateful firewall  →  it still serves

Why it still works. Cloud firewalls are stateful: responses to an allowed inbound connection flow back regardless of egress rules. So deny-all-egress plus allow-ingress is a complete request/response service that can never initiate. The model doesn’t fetch your data — your application pushes it in and receives the answer on the same connection. Retrieval data lives inside the enclave. Nothing reaches out.

Evidence, from inside the running model container on the sealed machine:

  • api.openai.com — no route. The address resolved; the packets went nowhere.
  • A raw IP with no DNS involved — no route. Rules out a DNS trick; the block is at the network layer.
  • Anthropic, HuggingFace, GitHub, raw ICMP — all unreachable.
  • The sanctioned bootstrap path answered normally, proving the enclave is operable rather than merely broken.
  • Inference served throughout.
The internet-bound packets produced no firewall deny log at all — they died at routing, before any rule was consulted. A firewall rule is a policy someone can weaken. An absent route is not a policy; it’s the absence of a thing.

The lockdown was so complete that Google’s own container agent couldn’t reach its registry until we explicitly allow-listed a private path for it.

The highest-trust move available

The hole we found in ourselves.

Our first seal leaked, and the way it leaked is instructive enough that we publish it.

Cloud instances resolve names through a link-local metadata service. It sits outside the egress firewall’s jurisdiction and performs recursive DNS resolution on the machine’s behalf. So even on a machine with no external address, no NAT, and a deny-all egress rule, a backdoor can encode stolen data into a hostname — <encoded-secret>.attacker.com — and the resolver dutifully walks the DNS hierarchy until the query reaches the attacker’s own authoritative nameserver. They read the data out of their query logs. Your firewall never sees a packet, because your machine never sent one. The metadata service did.

Bandwidth is roughly 30–60 bytes per query. Slow for bulk data; entirely sufficient for an API token or a credential.

How we proved it was live. We queried a hostname that had never existed, from inside the model container. It returned a definitive “domain not found” — not a timeout. A never-before-seen label can’t be answered from cache; a definitive answer is only possible if the query travelled the public hierarchy to that domain’s authoritative nameserver. The encoded label reached its operators. Demonstrated, not theorised.

How we closed it. The enclave initiates nothing, so it needs no resolver at all. We run the inference container with its resolver pointed at a dead in-network address, and the container runtime persists that setting — so the resolver is blackholed at every boot with no script to run and no ordering race to lose. Verified across a reboot — the test that caught two earlier attempts. Sealed and serving.

Why we publish our own failure. Our first evidence record noted that DNS resolved, but read it as strengthening the claim. The observation was honest; the interpretation was incomplete. Resolution is itself a channel. The corrected record supersedes the original on-chain, and both remain readable. An audit trail that can only record good news is not an audit trail.

02 — Dormancy · the audit surface

0

countable slots — not 120 billion floats.

The surface
36 × 128
Layers × experts, four routed per token. A Mixture-of-Experts gives you something discrete to count.
The outlier
0 fires
Each expert should see ~3% of tokens — tens of thousands of activations per million. One that never fires is screaming.
The check
Ablate
Disable the dormant set, re-run the benign evals. If quality doesn’t move, that capacity is dead weight — remove it.

Dormancy only catches a backdoor parked in unused capacity; a competent adversary would distribute it across hot experts. The closure comes from abandoning static analysis entirely — a backdoor must change the computation in order to change the output. In a Mixture-of-Experts, behaviour factors into which experts are selected and what they output; there is nowhere else for a trigger to live. So every request yields a routing fingerprint, and a trigger that reroutes is out-of-distribution — flagged without anyone knowing what the trigger is. Real routers specialise, so the output is a ranked suspicion list, not a verdict.

Where it runs

Your environment. Our perimeter.

The enclave runs inside your environment. It reads your data and has no path out. The control is cloud-agnostic — the same architecture in three dialects:

EnvironmentThe same control
GCPVPC + deny-egress firewall + private service access
AWSVPC + security groups + no internet gateway / no NAT + S3 endpoint
On-premphysical network + firewall

Who controls the egress? You do. In your account you hold root, and you could change the policy tomorrow. We won’t claim otherwise — and no serious security team would accept a vendor holding unilateral control of their network anyway.

What we provide instead is continuous proof. The policy ships as attested configuration; the harness re-verifies it and anchors the result; drift is detected and recorded. And because the sealed state contains zero allow-egress rules, there’s no rule to quietly weaken — defeating it requires adding a NAT gateway, which is a loud, auditable act that the next verification catches.

Verification beats control. You keep root. We keep receipts.
The one claim we won’t make

We will never tell you the model is clean.

Not because we haven’t looked — because nobody on earth can make that claim honestly, and the people who can approve your deployment already know it.

01

The weights are not code.

No source, no call graph, nothing to review. A backdoor would be an unknown input triggering unknown behaviour encoded across 120 billion numbers.

02

Nobody can rule it out.

Interpretability cannot prove the absence of a trigger at this size. Not us, not a lab, not a scanner. Anyone claiming otherwise is overclaiming.

03

So we built around it.

Four controls that don’t depend on the answer. Byte-identity, dormancy, containment, and the record hold whether or not anything is hiding in there.

The honest residual

What we do not claim.

This section is the point. Never fill the silence with a bigger claim — the residual is the credibility.

The answer channel is open. Zero egress does not close the semantic channel. A backdoored model can encode data in its reply to whoever asked. That reply goes to the caller — so an external attacker can’t read it unless the enclave is public-facing, in which case they could be a user, send a trigger, and receive data in the response. This is real. It’s bounded by recording every input/output pair and by routing-fingerprint anomaly detection. That is detection, not prevention. The Blackhole closes the network channel, not the semantic one.

Dormancy is a heuristic. A “dormant” expert may simply be specialised for a domain absent from the test corpus. Corpus breadth is load-bearing, and ablation — not the count — is the check that matters.

An adaptive adversary is not closed out. An attacker who shapes a trigger so its routing and activation signatures sit inside the benign distribution is not caught by our detectors. Nothing closes this. What changes is the cost: they must compromise the artifact the entire world runs, with a signature indistinguishable from normal computation, to produce an effect that cannot exfiltrate and is permanently recorded. That’s four independent controls deep — a research programme against a defended target, not an afternoon’s work.

Tiers are not interchangeable. Our fast tier runs on managed GPU infrastructure with open egress by design. We don’t claim containment there, and we never will. The sovereign tier is the one that carries the guarantee. Anyone conflating them is selling you something.

What is true today

Status, stated plainly.

A page that only lists green items is marketing. Where this says a thing isn’t proven yet, that’s not a roadmap — it’s the current truth.

CapabilityState
Egress lockdown, sovereign tierPROVEN — survives reboot; on-chain (block 1405)
DNS exfiltration channelCLOSED & verified across reboot
Tamper-evident audit trailLIVE
Two-layer access policyLIVE & enforcing
Byte-identity attestationVerified on real bytes; full-artifact run pending
MoE dormancy auditBuilt; not yet run against real weights
Runtime OOD scorerReference distribution emitted; scorer not built
120b served under containmentNOT YET — proven with a stand-in model. The control is model-agnostic: a firewall doesn’t care what serves.

The last row deserves emphasis. The containment proof used a small open model rather than the 120b, because the egress control is model-agnostic — proving one model can’t phone home and proving another can’t are the same proof. We say so rather than let the demo imply otherwise.

The implemented thesis

The application meets the substrate.

A substrate nobody builds on is a diagram. Hyfstele MLR reviews pharma promotional copy against the full FDA OPDP enforcement record — on a model that physically cannot phone home, with every decision on a tamper-evident chain. Your unreleased copy never leaves your perimeter, and you can prove it to your auditor.

Patent-pending PSI — US Prov. 64/069,537 SPRS 109/110 · CMMC L2-ready FIPS 180-4 signed records · 7-yr WORM