Bank Leumi, Israel’s largest bank, is relaunching its crypto trading service. The partner is not Paxos, as in 2022, but Galaxy Digital. The backbone is GK8—a custody platform that survived Celsius’s collapse. Code does not lie, but it can be misled. In this case, it carries the scars of a bankruptcy.
In July 2025, Bank Leumi announced a strategic partnership with Galaxy Digital to offer Bitcoin, Ethereum, and Solana trading through its Leumi Trade app and PEPPER digital bank. The service is slated for early 2027. The move comes after a 2022 attempt with Paxos was rejected by the Bank of Israel. This time, the bank chose Galaxy’s institutional-grade custody platform, GK8, which Galaxy acquired from Celsius’s bankruptcy proceedings in 2023 for $44 million. The acquisition included a 40-person team in Tel Aviv, led by GK8 co-founder Lior Lamesh. This is not a fresh start. It is a reboot on code that has already been battle-tested—and failed.
The architecture is a dedicated secure zone within the bank’s infrastructure. Customers execute trades without leaving the bank’s app. The assets are held in cold storage, managed by GalaxyOne trading platform. This is a classic centralized custody model. The security assumptions are clear: the bank holds the keys, and the customer trusts the bank’s operational security. Trust is a legacy variable. In DeFi, self-custody eliminates this variable. Here, it is reintroduced as a feature.
From a code perspective, GK8’s platform is a black box. No public audits of the current version are available. The original GK8 was audited by CertiK in 2022, but that audit covered a different deployment. The current integrated system with GalaxyOne has not been subjected to open peer review. Security through obscurity is not a strategy; it is a risk. Based on my experience auditing bZx v3 in 2020, I learned that even well-funded protocols can have hidden flaws. The same applies here. The difference is that bZx’s flaw was an integer overflow in a flash loan repayment. Here, the flaw could be a governance vulnerability in the custody layer.
The choice of Solana as a third asset is notable. Most bank-first crypto services start with BTC and ETH. SOL’s inclusion suggests Galaxy’s liquidity infrastructure in Israel covers SOL, and that institutional demand for Solana is rising. However, Solana’s history of network outages raises questions about the bank’s due diligence. ZK-circuits are compressing the future, but Solana’s proof-of-history is still a single point of failure. In my L2 scalability analysis, I observed that centralized bridges are the weakest link. Here, the bridge is the bank itself.
The market is treating this as a bullish signal for institutional adoption. I see a different story. The 2027 launch window is a two-year gap. During that time, regulatory frameworks will evolve. The Israel Capital Market Authority has proposed a draft allowing licensed firms to trade the top 50 digital assets. If that passes, Bank Leumi loses its first-mover advantage. The exclusive secure zone may become a commodity.
More importantly, the custody model replicates the same failure mode as Celsius: a single custodian holding large amounts of user funds. The only difference is that the custodian is now a bank, not a crypto lender. Banks are not immune to operational failures. The 2022 failure of the Paxos partnership was due to regulatory rejection, but the underlying security design was also questioned. This time, the design is more robust, but the centralization risk remains.
The acquisition of GK8 from Celsius is a red flag. Celsius’s bankruptcy revealed that GK8’s technology was not the problem—it was the governance. Code does not lie, but it can be misled by the people who control it. The same team now runs Galaxy Israel. Lior Lamesh is a competent engineer, but the organizational culture of a bankrupt entity may persist. In my cross-chain interoperability failure case study, I identified that centralized multi-sig wallets were the weakest link. Here, the multi-sig is the bank’s board.
Bank Leumi’s crypto service is a test case for the banking industry. If it succeeds, it will accelerate the narrative of banks as crypto gateways. If it fails—either through a security breach or regulatory blockage—it will set back institutional adoption for years. The real question is not whether the code is secure, but whether the trust placed in centralized custody is justified. Based on my experience auditing bZx v3 and analyzing L2 fraud proofs, I know that the weakest link is often the human layer. Trust is a legacy variable, and it is time to audit it.
The 2027 launch will be the moment of truth. Until then, watch the regulatory signals and the team’s operational security. The code may survive bankruptcy, but can it survive trust?