Worked example · 2022-12

Ankr deployer-key compromise and malicious upgrade — BNB Chain — 2022-12-01

Loss: approximately $5M direct extraction in the initial mint-and-swap event; downstream contagion via Helio Protocol, USDC swaps on PancakeSwap, and arbitrage against affected pools is commonly cited at $50M+ across all integrating protocols. Ankr publicly committed to making affected aBNBc / aBNBb holders whole and re-deployed under a new token contract.
OAK Techniques observed: OAK-T11.002 (Wallet-Software / Deployer-Key Compromise — former-employee credential reuse / authority-graph-not-rotated subclass) as the entry vector; OAK-T9.004 (Access-Control Misconfiguration — post-key-compromise) covering the ungated mint() capability that the malicious upgrade introduced, which the compromised key was then used to invoke; OAK-T5.003 (Hidden-Mint Dilution — the malicious upgrade introduced an unbounded mint() function with no access-control modifier, and the compromised key invoked it to mint a large quantity of aBNBc tokens from nothing, creating a counterfeit-issuance flow against the protocol's claimed token supply; the deployer-key-compromise + malicious-upgrade vector is the access-control-bypass variant of the broader hidden-mint class).
Attribution: inferred-strong — operator-attributed insider. Ankr's December 2 post-mortem identified the entry vector as a former team member who inserted a malicious code package via a supply-chain / social-engineering path that compromised the deploy key. This is a credible operator forensic attribution plus on-chain evidence of the insider-deployed contract, but the public record carries no named individual and no confirmed law-enforcement arrest — Ankr stated only that it was working with authorities to pursue prosecution — so it does not meet OAK's confirmed bar (official or multi-source attribution). The attribution applies to the entry-vector compromise; downstream extraction wallets remain pseudonymous in the public record.
Key teaching point: the underlying failure was not the contract code that was finally exploited but the authority graph that allowed a former employee's credentials to deploy and upgrade a production token contract. The case pairs directly with Wintermute (/examples/2022-09-wintermute.md) on the rotate-on-disclosure / authority-graph-enumeration lesson — Wintermute did not rotate keys after a public vulnerability disclosure; Ankr did not rotate authority after an employee separation.

Summary

Ankr is a multi-chain infrastructure provider whose product set in 2022 included liquid staking on BNB Chain, with the aBNBc and aBNBb tokens representing staked-BNB positions. On 2022-12-01, an attacker used a deployer key controlling the aBNBc token contract to publish a malicious upgrade to the contract's logic. The upgrade introduced an unbounded mint() capability with no access-control modifier; the compromised key then invoked that capability to mint a large quantity of aBNBc out of nothing.

The minted aBNBc was sold across PancakeSwap and other BNB Chain venues, draining liquidity and crashing the aBNBc price. Downstream protocols that had treated aBNBc as collateral — most notably Helio Protocol, where the price drop allowed an attacker to acquire substantial HAY stablecoin debt against effectively-worthless aBNBc collateral — absorbed the secondary damage. Cross-chain swaps and arbitrage against the broken peg propagated losses to additional integrating protocols. Industry estimates of total contagion across the affected ecosystem reached $50M+, although the direct on-Ankr extraction was approximately $5M.

Ankr paused the affected contracts, revoked the compromised authority, and within ~36 hours had identified the entry vector publicly as a former employee whose credentials had retained access to deployment infrastructure after their separation from the firm. In January 2023, the Republic of Korea announced an arrest of the former employee in connection with the incident — making this one of the comparatively rare 2022 DeFi-class incidents with a named-individual attribution backed by a public criminal-justice record.

For OAK's purposes, this incident is the canonical 2022 case study for the compromised key → malicious upgrade → unbounded mint chain, where the entry vector is operational (employee credential lifecycle) rather than cryptographic (insufficient entropy, signing-vendor compromise, smart-contract bug).

Timeline (UTC)

When	Event	OAK ref
Pre-event (timing not publicly precise)	Former-employee credentials retain access to Ankr's deployer-key authority on BNB Chain after separation	(standing T11.002 surface)
2022-12-01 (single transaction)	Compromised deployer key publishes a malicious upgrade to the aBNBc token contract; upgrade introduces an ungated `mint()` capability	T11.002 + T9.004 setup
2022-12-01 (same window)	`mint()` invoked from the compromised key; large quantity of aBNBc minted out of nothing	T11.002 + T9.004 extraction
2022-12-01 (within minutes)	Minted aBNBc sold across PancakeSwap and other BNB Chain venues; aBNBc price crashes	T5 outflow
2022-12-01 (same day)	Secondary attackers exploit the aBNBc price crash against Helio Protocol — acquire HAY debt against effectively-worthless aBNBc collateral	(downstream contagion)
2022-12-01 to 2022-12-02	Ankr pauses affected contracts; revokes compromised authority; begins re-deployment under a new aBNBc contract	(defender response)
2022-12-02	Ankr publishes post-mortem identifying the former-employee entry vector	(attribution)
2023-01 (announced)	Republic of Korea law-enforcement arrest of the former employee in connection with the incident	(criminal-justice action)

What defenders observed

Pre-event (authority-graph layer): Ankr's deployer-key authority over the aBNBc token contract was held by a credential that had not been rotated after the relevant employee's separation from the firm. The standing surface here is not "the deployer key was compromised" but "the deployer key's authority graph was not enumerated and rotated as part of the offboarding process." This is the load-bearing operational artefact for the case.
At-event (upgrade-publication signal): the malicious upgrade was published on-chain. A monitoring path that watched for unscheduled bytecode changes on production token contracts — and especially for upgrades that introduced new external entry points such as an ungated mint() — would have produced a high-confidence signal at the moment of publication, before any tokens were minted. The interval between upgrade publication and first malicious mint() was short, but it was non-zero and was on-chain-observable.
At-event (mint-and-swap signal): the unbounded mint followed by immediate large-scale sell pressure on PancakeSwap is a recognisable on-chain pattern. Cross-protocol monitoring of liquid-staking-token / DEX-pool ratio invariants would have produced a second independent signal as the aBNBc price diverged from the underlying staked BNB.
Post-event (attribution): Ankr's same-week public identification of the entry vector as a former employee, followed by the publicly-announced criminal-justice action, is unusual for the 2022 cohort and gives this case a stronger attribution-strength rating (named-individual / authoritative criminal-justice action) than most adjacent incidents.

What this example tells contributors writing future Technique pages

T11.002 covers operational authority lifecycles, not just key-extraction events. Wintermute (/examples/2022-09-wintermute.md) is the canonical case for cryptographic key-generation weakness; Ankr is the canonical case for operational authority lifecycle. Both are T11-class compromises but the failure modes are independent: Wintermute's key was correctly held but cryptographically weak; Ankr's key was cryptographically strong but held by a separated employee. T11 contributors should describe the surface at the level of "authority graph at the moment of the incident" rather than at the level of "the key value itself."
T9.004 here is post-key-compromise contributory. The ungated mint() capability that the malicious upgrade introduced is a real T9.004 surface, but it would not have existed without the upgrade publication, which would not have occurred without the compromised deployer key. The chain is T11.002 → T9.004 → extraction. Worked examples should describe the chain rather than collapsing it onto whichever Technique looks most visible.
The upgrade-publication moment is a load-bearing detection surface. OAK's detection guidance for T11.002 should explicitly include monitoring for unscheduled bytecode changes on production contracts and for newly-introduced privileged external entry points (especially mint(), burn(), pause(), setOwner(), upgradeTo()). The window between upgrade publication and first malicious invocation is short but non-zero in many T11.002 cases; defenders that monitor at upgrade-publication granularity get a head-start that defenders that only monitor at extraction-event granularity do not.
Cross-protocol contagion belongs in the worked example. The direct on-Ankr extraction was ~$5M; the cross-protocol contagion through Helio and PancakeSwap exceeded that figure by an order of magnitude. Worked examples for T11-class compromises that affect liquid-staking tokens, stablecoins, or other widely-composed assets should track the cross-protocol blast-radius explicitly. The Euler worked example (/examples/2023-03-euler-finance.md) makes the same observation for a T9.004 incident.
Named-individual attribution backed by a criminal-justice action is a v0.1 attribution-strength category that few cases meet. Ankr meets it; Mango Markets meets it (Eisenberg, with the 2025 vacatur nuance); most other 2022 DeFi-class incidents do not. T11 contributors should record attribution strength explicitly and avoid promoting industry speculation to inferred-strong.

Public references

Ankr — Ankr Hack: Post-Mortem and Way Forward — Ankr's own post-mortem identifying the former-employee entry vector.
BlockSec — aBNBc Exploit: Root-Cause Analysis — function-level walkthrough of the malicious-upgrade and mint() chain.
PeckShield — Ankr aBNBc Exploit Analysis — same-day on-chain trace of the upgrade and mint events.
Halborn — Explained: The Ankr Protocol Exploit (December 2022) — defender-oriented technical post-mortem.
The Block — Ankr says hack on aBNBc was caused by a former employee — coverage of Ankr's attribution statement.
Cointelegraph — Ankr exploiter arrested in South Korea — coverage of the January 2023 arrest announcement.
[zhou2023sok] — academic taxonomy covering this as an operator-side custody-surface compromise feeding a contract-upgrade chain.

Discussion

Ankr is OAK's canonical 2022 case for the operational-authority-lifecycle subclass of T11.002. The case pairs deliberately with Wintermute (/examples/2022-09-wintermute.md) in the v0.1 example set: both are T11-class compromises, both have public attribution and post-mortems of unusually high quality for the cohort, and both turn on the same operational-discipline question that OAK contributors should be drilling into in T11 pages — at the moment of the incident, what was the authority graph, and was it the authority graph the protocol thought it was?

For Wintermute, the authority graph included an admin-vault privilege held by a Profanity-generated vanity address whose entropy was insufficient and which had not been rotated after the public 1inch disclosure. For Ankr, the authority graph included a deployer-key privilege held by a credential that had not been rotated after the relevant employee's separation from the firm. The cryptographic shape is different; the operational shape is identical: a privilege held by a key that should not have been in production by the time of the incident.

OAK v0.1's T11.002 page should be written so that both cases land cleanly under it. The v0.x candidate split — separating cryptographic key-generation weakness (Wintermute) from operational authority-lifecycle (Ankr) into different sub-Techniques — is a real taxonomy question, and the choice will turn on whether the framework prioritises the entry-vector mechanism (cryptographic vs operational) or the operational-discipline lesson (rotate-on-disclosure / authority-graph-enumeration). Either choice is defensible; OAK contributors should make the choice deliberately rather than letting the taxonomy drift on this axis.

The Chainalysis 2022 retrospective places insider / operational-credential incidents at a smaller absolute count than smart-contract-side incidents but at a higher per-incident loss when they occur, because the entry vector typically reaches into root-of-trust authority rather than peripheral state. Ankr fits that cohort precisely; the 2023 arrest announcement is the unusually clean attribution endpoint that most cohort members do not reach.