The Radix Foundation disclosed on September 17 that a vulnerability introduced during a 2023 Radix Engine code refactor was exploited on August 31, allowing attackers to drain about $1.26 million from multiple vaults. Because the flaw affected the Radix Engine, which executes transactions and verifies asset ownership, Radix validators temporarily stopped network consensus and blockchain transactions were suspended for more than 10 days.
The incident extended beyond the assets directly stolen. Investigators said the vulnerability could theoretically have been used against any vault on the network, including tokens held in user accounts, applications and liquidity pools. For trading platforms and liquidity providers, the central issue was that the execution layer failed to enforce the permission boundaries designed to protect vault assets.
26 transactions moved multiple assets
A reconstruction of the community ledger found that the attacker carried out 26 exploit transactions between 16:02 and 16:57 UTC on August 31. The transactions moved approximately 458,915 USDC, 72,420 USDT, 61.08 ETH, 6.35 Wrapped Bitcoin, 536.16 SOL and 32.91 BNB. Based on market prices on August 31, the assets were worth about $1.26 million in total, with the two stablecoins accounting for approximately $531,300.
The attacker also withdrew 13,000 XRD from one vault to pay transaction fees. The stolen assets were then transferred through Hyperlane to Ethereum, BNB Chain and Solana, where they were exchanged for ETH. The Radix Foundation said the Hyperlane bridge operated as designed. The attacker first obtained the assets through the Radix Engine and only then used the cross-chain service to move the funds. There is currently no indication that private keys were compromised.
The amount transferred may not capture the full economic impact of the incident. Investigators said the flaw was not limited to the cross-chain assets directly withdrawn by the attacker; other tokens and assets held in vaults across the network could also have faced the same risk.
A code refactor introduced the flaw in 2023
The Radix Foundation said the defect was introduced by the RDX Works development team in June 2023, while cleaning up Radix Engine code. The vulnerability remained in the network for more than three years before being exploited on August 31, 2026.
The problem involved how the engine handled vault references. A transaction could use an internal address to point to another user’s vault and pass that reference to a specific smart contract. Radix Engine would then allow a standard withdrawal function to be called without correctly enforcing the vault’s ownership check, meaning it failed to reject an unauthorized request.
Under the intended design, only a vault owner or an authorized transaction can withdraw its assets. The defect potentially allowed an attacker to access funds in user accounts, applications and liquidity pools without obtaining a signature from the asset owner.
The code had also passed an independent security review. Zellic audited the Radix protocol in 2024, including the engine core where the defect was located, but did not identify the authorization vulnerability. Because the flaw originated in an earlier code refactor, the incident highlighted how routine maintenance can alter existing security assumptions in ways that conventional audits may not detect.
Validators suspended consensus to stop further transactions
Once investigators confirmed that the problem lay in the transaction execution layer rather than in a single application, Radix validators coordinated a response. They voluntarily took enough stake offline to prevent the network from reaching consensus and processing additional transactions.
The shutdown lasted more than 10 days. The protocol fix added checks preventing restricted vault references from being used with standard withdrawal functions. A reconstruction of the community ledger indicates that user transactions resumed on September 11.
The incident also produced secondary losses in liquidity pools. After the attacker removed cross-chain assets from one side of several trading pairs, the pools’ price ratios became distorted. Another account used the abnormal prices to withdraw millions of XRD from affected pools. The episode showed that even after the initially stolen assets have left the network, an execution-layer vulnerability can continue to generate losses through price distortion and changing liquidity conditions.
Radix to expand regression testing and emergency procedures
The Radix Foundation said its follow-up measures will include expanding regression tests, strengthening security-review procedures and formally institutionalizing the emergency actions validators took to restore network security. The work will focus not only on closing the exploited code path, but also on reviewing all code changes that affect authorization logic.
The Foundation also said future security work must account for increasingly capable AI-assisted code-analysis tools, which it believes could help attackers locate long-standing defects. The disclosure did not establish whether the attacker used any particular tool.
For developers and validators, the immediate task is to ensure that routine maintenance does not alter the permission model without sufficient testing and review. For users and liquidity providers, the key concerns remain whether the network consistently enforces vault ownership checks and whether the revised emergency procedures can be activated quickly in a similar incident. Those factors will shape confidence in the security of Radix assets and the stability of the network.