Matter Labs publishes the permissioning code the Bundesbank is running

What happened

Matter Labs, the company that develops ZKsync, announced on 8 September 2026 that it was publishing the permissioning engine of Prividium, its distributed ledger platform for financial institutions, as open source software, and named the Deutsche Bundesbank as the first institution to deploy the platform. The announcement says the decision 'addresses a requirement that defines how financial institutions adopt new infrastructure: the ability to run, inspect, and modify the technology they operate, independently of any single vendor', and describes the Bundesbank's involvement precisely: Germany's national central bank 'has deployed the platform self-hosted in its own infrastructure, where smart contract and token data remains within its own environment, and the two organisations are collaborating on its design and testing'. Alex Gluchowski, the chief executive, is quoted: 'Financial institutions should not have to choose between adopting new infrastructure and keeping control of it. Publishing the Prividium permissioning engine as open source means an institution can run a permissioned chain from public code, in its own environment. Inspect it, modify it, and reduce its reliance on any single vendor while settling on a shared, credibly neutral layer. The Bundesbank is the first institution to deploy the platform this way, and we are glad to be collaborating with them on its design and testing.' A release note published the same day sets out what was actually shipped. Prividium Core is 'one application and six packages': a permissions API described as 'the system of record for roles, access controls, and permission decisions on a Prividium chain', the policy and role primitives beneath it, a client SDK, shared server and database helpers, a generated TypeScript client, instrumentation, and test dependencies. The licence is Apache 2.0 and the repository is github.com/matter-labs/prividium-core. Matter Labs says it is 'a working codebase, not a snapshot: it is the same code we develop and run, published byte-for-byte', mirrored from an internal monorepo by git subtree and synced in both directions, with public continuous integration and contributions merged in the public repository first. It also says 'Public history starts at a single commit', because the monorepo's history does not travel with the mirror. What remains commercial is listed: 'The administration console, user access panel, and connectors (off-chain integrations into core banking systems, custody, ISO 20022 messaging, and enterprise logging) remain commercial products, delivered under agreement along with enterprise support, certified builds, and signed releases for institutions moving to production.' The claim the reports have taken up is that the released code is enough on its own. Under a heading reading 'It runs standalone', the release note says Prividium Core 'is designed to run against the already open-source ZKsync OS stack' with the Atlas sequencer, the Airbender prover, the interoperability contracts and the block explorer components alongside, and that 'Cloning Prividium Core and running it against ZKsync OS is designed to yield a functioning permissioned chain: roles, access controls and permissions enforced by the engine, execution and proving by ZKsync OS, correctness verifiable through ZK proofs anchored to Ethereum. No commercial component is required for that path.' The formal announcement is more guarded about the same thing, saying the engine 'is planned to run standalone, so that an institution will be able to operate a permissioned chain from public code, in its own environment, without a commercial agreement', and a CEO essay published on 9 September repeats that it 'is designed to run standalone'. None of the three says the path has been walked. The repository confirms what was published and adds two things the announcements do not. Its README describes Prividium Core as 'the permissioned RPC proxy and permissions API behind a Prividium network', which 'authenticates callers, applies role-based permissions to JSON-RPC and REST requests, and forwards what it allows to a ZKsync sequencer', and states in terms that 'It is one component of the Prividium product. It is not the full commercial offering'. The same file reserves the name: 'Prividium is a registered trademark of Matter Labs. Use of the Prividium name for modified or redistributed versions requires written approval from Matter Labs.' The repository was created on 7 September 2026 at 09:45 UTC, the day before the announcement, was last pushed to on 10 September, and on 13 September carried no stars, one fork and eight open issues. The licence file is the Apache License 2.0 reproduced in full beneath a Matter Labs copyright line; GitHub's own API classifies it as NOASSERTION, which is an artefact of the preamble rather than a departure from the licence. The Bundesbank has said nothing. The newest item on its English press release index is dated 13 August 2026, and no release, speech or note about the deployment could be found on its site. Everything known about the central bank's involvement comes from the vendor.

Why it matters

The argument Matter Labs is making is the one this beat has been circling for three years, and it is not really about cryptography. Regulated institutions have spent a decade being asked to run money on infrastructure they cannot inspect, cannot leave and do not control, and they have mostly declined. Gluchowski's essay puts the analogy on the table directly: open-weight models exist in artificial intelligence because states and enterprises refused to build on infrastructure they could neither inspect nor run themselves, and 'a bank can tolerate a vendor dependency in its CRM. It cannot tolerate one in the system that moves money.' For a central bank the standard is higher still, which is why a central bank running the code is the proof point the company wanted. Whether the substance matches the framing depends on what was published, and the repository is the place to check. What is now public is the layer that decides who may read and write, and it sits in front of a sequencer rather than being one. That is a real thing to open: on a permissioned chain, the permissions are the policy, and an operator that cannot read the code enforcing them is trusting a vendor with exactly the decision it is not allowed to delegate. But it is narrower than 'the core of Prividium', which is what the announcement's own title calls it, and it leaves the commercially load-bearing pieces, the administration console, the access panel and the connectors into core banking and ISO 20022 messaging, on the other side of a contract. An institution that reads the permissioning code and still cannot operate without the console has inspected the lock and rented the door. The second thing worth seeing is what open source does and does not remove. Apache 2.0 with a trademark reservation is a conventional and defensible arrangement, and it is the same shape the rest of the industry uses; but 'sovereignty that depends on a vendor's goodwill is not sovereignty', which is the essay's line, sits oddly beside a requirement for written approval to use the name on a modified build. The dependency that matters is not the name, though. It is that the permissioning engine forwards to a ZKsync sequencer, so the chain an institution runs from public code is still a ZKsync chain, anchoring to Ethereum, with the economics and upgrade path of that stack. Third, this is the privacy question that has held institutional adoption back, answered in the only way institutions have ever accepted: keep the data inside, publish a proof that the arithmetic was right. It is the same design the corpus has been tracking through the ECB's work on a digital capital market and through the shielded-pool research in the library, and the Bundesbank's participation, if it is what the vendor says it is, is the strongest signal yet that a G7 central bank regards the trade as acceptable. Which makes the silence from Frankfurt the most important fact in the story. Every word about what the Bundesbank is doing, why, and on what scale comes from the company selling the software. A test is not an endorsement, a deployment in a lab is not a payment system, and a central bank that has not published a line about a project can walk away from it without ever having to explain. The corpus has seen exactly this before with vendors announcing central bank engagements that turned out to be smaller than the press release implied. Until the Bundesbank writes something down, the right reading is that Matter Labs has published its code and told the world who is looking at it.

What is not settled

Nobody has demonstrated the standalone path. All three documents hedge it in the same way: the announcement says the engine 'is planned to run standalone', the release note says cloning it and running it against ZKsync OS 'is designed to yield a functioning permissioned chain', and the essay says it 'is designed to run standalone'. None of them reports that anyone has done it, and the reports covering the announcement drop the hedge and say an institution can now run a permissioned chain from public code. The repository is too new for the outside world to have settled the point: created on 7 September, no stars and one fork on 13 September, eight issues open. What the Bundesbank is actually testing is unknown. It has published nothing, its English press release index has carried nothing since 13 August 2026, and the vendor's description, self-hosted with smart contract and token data inside its own environment, does not say which of the bank's functions the test touches, whether anything settles across it, whether any counterparty is involved, how long it runs or what would count as success. It also does not say whether the bank is running the published code or a build supplied under agreement, which is the distinction the whole announcement turns on. The line between open and commercial is drawn by the vendor and can be redrawn by the vendor. The administration console, the user access panel and the connectors into core banking, custody, ISO 20022 messaging and enterprise logging are named as remaining commercial, and the release note asserts that 'The open core is complete without them', which is a judgement rather than a demonstration; the README is more cautious, saying Prividium Core 'is not the full commercial offering'. Nothing in the licence stops Matter Labs moving a future component the other way. The claim that the repository holds 'the same code we develop and run, published byte-for-byte' cannot be checked from outside, because the public history starts at a single commit and the monorepo it is mirrored from is private, so an observer can verify what is there but not that nothing was removed on the way out. The trademark reservation is undefined at the edges: written approval is required to use the Prividium name for modified or redistributed versions, and nothing says what a fork may call itself or on what terms approval is given or withheld. And the question the CEO's essay raises and leaves open is the one that decides whether any of this becomes a market: how chains run by individual institutions reach each other and reach public markets without giving up the control that made them attractive. He calls it 'the foundation, not the finish line' and says the answer is coming, which is a promise rather than a design.