The Radix Foundation has published a public incident report for the August 31 exploit that drained roughly $1.3 million from the network, tracing the root cause to a vault-authorization flaw introduced during a June 2023 code cleanup — a defect that lived in production for more than three years, survived an independent security audit, and was only contained after validators deliberately halted consensus for more than ten days.

The report, published September 17 and corroborated by CryptoSlate, CoinPaprika and KuCoin News, is a rare full-disclosure document from a layer-1 foundation: it names the bug's origin, the audit that missed it, the exact transaction window of the theft, and the emergency mechanism that stopped it.

How the Flaw Worked

According to the report, a development team at RDX Works — the company behind the Radix protocol — introduced the defect during routine tidying of the Radix Engine, the software layer that executes transactions and enforces asset ownership across the network.

The bug changed how the engine handled vault references, the containers Radix uses to hold tokens inside accounts and applications. A transaction could identify another user's vault by its internal address and pass that reference into purpose-built smart-contract code. The engine then allowed ordinary withdrawal functions to be called without enforcing the ownership boundary that should have rejected the request.

The practical effect: an attacker could withdraw assets held by user accounts, applications and liquidity pools without ever obtaining the owners' signatures. No private keys were compromised at any point in the incident, the Foundation said.

The Theft: 26 Transactions in 55 Minutes

A community reconstruction of the ledger cited by the Foundation shows the attacker completed 26 exploit transactions between 16:02 and 16:57 UTC on August 31. The withdrawals covered six bridged assets: approximately 458,915 USDC, 72,420 USDT, 61.08 ETH, 6.35 wrapped Bitcoin, 536.16 SOL and 32.91 BNB — worth roughly $1.26 million at same-day prices. The attacker also took 13,000 XRD, the network's native token, to pay transaction fees.

The assets were moved out through the Hyperlane bridge to Ethereum, BNB Chain and Solana, then sold for ETH. The Foundation noted that Hyperlane itself operated as designed throughout: the attacker had already obtained the assets through the Radix Engine before using the bridge as an exit, not an entry.

The Halt: Validators Broke Liveness on Purpose

Once investigators determined the vulnerability sat in the execution layer rather than a single application, validators coordinated to stop transaction finalization — taking enough staked shares offline to prevent the network from reaching consensus. The deliberate liveness break occurred between roughly 20:30 and 23:30 UTC on August 31, hours after the exploit began.

The halt lasted more than ten days. A protocol fix added checks preventing a restricted vault reference from being used for an ordinary withdrawal, and user transactions resumed on September 11, per the community ledger reconstruction.

The Foundation's assessment of what was at stake justifies the drastic step: investigators concluded the flaw could have been used against any vault on the network, not only the bridged holdings the attacker targeted. The attacker chose bridged assets because they were easiest to exit through external liquidity.

The Audit That Missed It

The most uncomfortable fact in the report is that the flaw had already been through an independent human review. Zellic audited the Radix protocol in 2024 — including the engine kernel containing the defect, which had been introduced the previous year — and did not detect the authorization flaw. CoinPaprika notes the engagement passed the code with a high score.

The miss is a data point in an ongoing industry argument about audit scope and shelf life: a review of a complex execution engine apparently did not — or could not — re-derive the authorization semantics of every code path affected by a routine refactor. A 2023 maintenance change silently altered a security assumption the audit examined the next year. TrustGrade News covered related research this week finding that 94% of DeFi losses occur in code outside audited scope; Radix is the harder case — the loss occurred in code that was inside scope, after the assumption underneath it moved.

Secondary Damage and the AI Question

The incident report also documents second-order losses: after the attacker removed bridged assets from one side of trading pairs, the distorted prices allowed another account to extract millions of XRD from affected liquidity pools. An execution-layer failure kept causing economic damage even after the initial assets left the network.

The Foundation also said future security work must account for increasingly capable AI-assisted code-analysis tools, which it believes may have helped the attacker identify the years-old defect. That belief is stated as an assessment, not a finding — no forensic evidence identifying the attacker or their methods has been published, and the attribution question remains open.

Remediation listed in the report includes regression tests for the patched behavior, a strengthened security review process around changes to authorization logic, and formalization of the emergency procedure validators improvised in August. The patched code has since been independently reviewed, the Foundation said.

TrustGrade covers the security and trust ecosystem around digital assets. For verified trust data on the platforms and firms shaping it, see trustgrade.ai.