OAK — OnChain Attack Knowledge

Worked example · 2025-09

SwissBorg SOL-Earn third-party-API supply-chain compromise — Solana — 2025-09-08

Loss
approximately $41M (~192,600-193,000 SOL tokens at the spot price at the time of the event) extracted from SwissBorg's SOL Earn staking program on Solana on 2025-09-08. SwissBorg is a Switzerland-based crypto investment / wealth-management platform offering retail users access to staking and yield-earning products; the SOL Earn program was the affected surface, with the broader SwissBorg platform across other earning products and assets unaffected. The breach impacted approximately 1% of SwissBorg's user base and approximately 2% of total assets under management.
Recovery
partial via white-hat-coordinated cross-venue blocking + operator-funded user-side compensation. SwissBorg engaged with white-hat hackers, blockchain investigators, and partners including Fireblocks and the Solana Foundation to track stolen funds; transactions linked to the exploit were blocked on multiple exchanges before they could be cashed out, materially limiting attacker-side liquidity realisation. Affected users were compensated from the SwissBorg treasury under operator-side reimbursement, producing a near-100% user-side recovery via the operator-team-reimbursement primitive. Net protocol-treasury cost approximated the un-recovered on-chain portion plus the cross-venue-blocked-but-not-recovered portion.
OAK Techniques observed
OAK-T11.001 (Third-Party Signing / Custody Vendor Compromise) — primary, in the third-party-staking-API sub-vector: the breach was caused by compromise of a third-party API at SwissBorg's staking partner Kiln, not by compromise of SwissBorg's own infrastructure. OAK-T15.001 (Social Engineering of Operator Personnel) — anchor for the partner-API / fake-staking-partner pretext sub-shape; the attacker leveraged the partner-trust relationship between SwissBorg and Kiln as the trust-substrate for the compromise. The compromise vector is structurally analogous to the WazirX July 2024 case at /examples/2024-07-wazirx.md (Liminal Custody compromise) and the DMM Bitcoin May 2024 case at /examples/2024-05-dmm-bitcoin.md (Ginco wallet-software-vendor compromise) — all three are 2024-2025 worked examples of the same OAK-T11.001 third-party-vendor-compromise umbrella with sub-vector variation across custody-vendor / wallet-software-vendor / staking-API-vendor. OAK-T9.004 (Access-Control Misconfiguration) is the on-chain extraction-mechanism sub-pattern: the attacker performed a standard unstaking transaction that concealed malicious logic several days before the attack was performed, with eight authorisation instructions designed to transfer control over several of the platform's staking accounts from SwissBorg to attacker-controlled on-chain accounts. The two-phase pattern (delayed-trigger malicious instructions + later activation) is operationally distinctive.
Attribution
pseudonymous but not OAK-G01-confirmed as of v0.1. The targeted vendor-side compromise of a staking-infrastructure provider (Kiln) and the multi-instruction delayed-trigger malicious pattern share structural features with the OAK-G01 / Lazarus-aligned 2024-2025 wave (DMM Bitcoin, WazirX, Radiant Capital, Bybit), but no public technical indicator — wallet-cluster correlation, on-device forensics, or FBI / Treasury / Mandiant attribution — links the SwissBorg incident to a specific OAK-G01 cluster as of OAK v0.1. Contributors writing future attribution updates should monitor for forensic-provider-led cluster correlation that may move the attribution status.
Key teaching point
the third-party-staking-API supply-chain surface is an OAK-T11.001 sub-vector that the 2024-2025 corpus had not previously documented at the protocol-and-platform-integration layer. The DMM Bitcoin (wallet-software-vendor), WazirX (custody-vendor), and Radiant Capital (developer-laptop) cases had documented OAK-T11.001 sub-vectors covering vendor-side wallet-software supply chain, vendor-side custody platforms, and operator-side developer workstations. The SwissBorg case adds the third-party-staking-infrastructure-API sub-vector as a fourth principal sub-vector under OAK-T11.001. Defender / Mitigations-layer practice should now treat staking-infrastructure providers (Kiln, Figment, Stakefish, Allnodes, etc.) as discrete OAK-T11.001 target-surface categories alongside custody vendors (Liminal, Fireblocks, BitGo, etc.) and wallet-software vendors (Ginco, MetaMask, Trust Wallet, etc.). The cross-vendor-category coverage of the OAK-T11.001 surface is now broad enough to support a Mitigations-layer cohort entry for third-party-vendor-supply-chain threat-modelling that spans all four sub-vectors.

Summary

SwissBorg is a Switzerland-based crypto investment / wealth-management platform serving retail users with staking, yield-earning, and asset-management products. The platform's SOL Earn program offered users a staking-backed yield product on Solana SOL tokens, with the underlying staking infrastructure operated by Kiln as a third-party staking-API provider. Kiln operates similar staking-infrastructure services for multiple platform-and-protocol customers across the broader DeFi / CeFi ecosystem.

On 2025-09-08, an attacker exploited a compromised third-party API at Kiln to execute an unstaking transaction against SwissBorg's SOL Earn program that drained approximately 192,600-193,000 SOL tokens (~$41M at the spot price). Per the SwissBorg post-incident statement and the corroborating Halborn post-mortem, the attacker had submitted eight authorisation instructions concealed within a standard unstaking transaction several days before the actual extraction, designed to transfer control over several of the platform's staking accounts from SwissBorg to attacker-controlled on-chain accounts. When the actual unstaking occurred, the previously-injected authorisation instructions activated and the staking account control was transferred, with the SOL drained to attacker-controlled addresses.

SwissBorg's response chain ran rapidly within hours of detection. The platform engaged with white-hat hackers, blockchain investigators, and ecosystem partners including Fireblocks (custody / institutional-asset-management infrastructure) and the Solana Foundation (chain-level coordination authority on Solana). Transactions linked to the exploit were blocked on multiple exchanges before they could be cashed out, materially limiting attacker-side liquidity realisation. Affected users were compensated from the SwissBorg treasury under operator-side reimbursement, producing a near-100% user-side recovery.

For OAK's purposes the case is a clean OAK-T11.001 third-party-staking-API supply-chain extraction with on-chain T9.004 multi-instruction-delayed-trigger as the extraction mechanism. The novel OAK contributions are: (a) documenting the third-party-staking-API sub-vector of OAK-T11.001 as distinct from the custody-vendor (WazirX), wallet-software-vendor (DMM Bitcoin), and developer-laptop (Radiant Capital) sub-vectors; (b) capturing the delayed-trigger malicious-instruction-injection pattern as an on-chain extraction sub-pattern with operational distinctiveness from same-transaction direct-call patterns; and (c) capturing the multi-actor cross-venue-coordinated blocking as a recovery primitive that produces partial recovery via attacker-side-liquidity-realisation friction rather than via on-chain unwinding.

Timeline (UTC)

When Event OAK ref
Pre-event SwissBorg deploys SOL Earn program on Solana with Kiln as third-party staking-API provider; Kiln's API surface is integrated with SwissBorg's user-side staking-account management (T11.001 surface — third-party-staking-API integration)
Pre-event Attacker compromises Kiln's API surface via undisclosed off-chain means; precise compromise vector not exhaustively detailed in the public record T11.001 surface (broadly construed)
Pre-event (several days before extraction) Attacker submits eight authorisation instructions concealed within a standard-appearing unstaking transaction, designed to transfer staking-account control from SwissBorg to attacker-controlled on-chain accounts on subsequent activation (delayed-trigger malicious-instruction-injection setup)
2025-09-08 (attack window) Attacker activates the previously-injected authorisation instructions; staking-account control transferred to attacker-controlled accounts; 192,600-193,000 SOL ($41M) drained T11.001 + T9.004 extraction
2025-09-08 (within hours) SwissBorg detects anomalous staking-account state; pauses affected program functionality; engages white-hat hackers and blockchain investigators (operator response)
2025-09-08 onward SwissBorg coordinates with Fireblocks (custody / institutional-asset-management partner) and the Solana Foundation (chain-level authority); cross-venue blocking of attacker-linked transactions on multiple exchanges (white-hat-coordinated cross-venue blocking)
2025-09-08 / 2025-09-09 SwissBorg publishes public acknowledgement; identifies Kiln third-party API breach as the entry vector; commits to user-side full reimbursement from SwissBorg treasury (operator response / transparency)
2025-09 onward Industry detection chain: Halborn publishes defender-oriented post-mortem; PeckShield issues cumulative-loss aggregation; SlowMist attaches funding-source tracking; Cyvers corroborates the multi-instruction delayed-trigger pattern (multi-investigator coverage)
2025-09 onward User-side reimbursement from SwissBorg treasury proceeds in stages; near-100% user-side recovery achieved (operator-team reimbursement recovery primitive)
2025-09 onward Kiln publishes its own incident-response statements covering the third-party API surface compromise and remediation steps (vendor-side response)
Continuing Pseudonymous attribution; no civil-forfeiture action; no named-individual claim; no public OAK-G01 cluster correlation as of v0.1 (recovery state)

What defenders observed

  • The third-party-staking-API supply-chain surface is operationally distinct from the third-party-custody and third-party-wallet-software surfaces. Kiln operates a staking-API service that integrates with multiple platform-and-protocol customers across DeFi / CeFi; the surface's structural property is that the API-provider holds operational authority to construct staking transactions on behalf of customer platforms. Compromise of the API surface produces transactions that customer platforms execute as if legitimate — analogous to but operationally distinct from custody-vendor compromise (where the vendor holds key authority directly) and wallet-software-vendor compromise (where the vendor produces signed transactions via compromised software). Defender / Mitigations-layer practice should treat staking-API providers as a discrete fourth third-party-vendor sub-vector under OAK-T11.001, with cohort-level monitoring of the staking-infrastructure-provider category (Kiln, Figment, Stakefish, Allnodes, P2P, Chorus One, etc.).
  • The delayed-trigger malicious-instruction-injection is an operationally-distinctive on-chain pattern. The attacker submitted eight authorisation instructions concealed within a standard-appearing unstaking transaction several days before the actual extraction; the malicious instructions activated only on subsequent unstaking events. Defender / on-chain-monitoring practice for staking-account-control surfaces should include detection of (a) authorisation-instruction state divergence between current and prior transaction states, (b) staking-account ownership changes outside of expected upgrade / migration windows, and (c) anomalous instruction patterns within otherwise-standard unstaking flows. The Solana account-model architecture makes this category of analysis structurally feasible; the defender / Mitigations-layer lesson is that the analysis is not yet uniformly deployed across staking platforms.
  • The multi-actor cross-venue-coordinated blocking is a working partial-recovery primitive. SwissBorg's coordination with Fireblocks and the Solana Foundation produced cross-venue transaction blocking on multiple exchanges before the attacker could cash out. The blocked transactions did not unwind on-chain extraction but did limit attacker-side liquidity realisation. Defender / Mitigations-layer practice should treat cross-venue-coordinated blocking as a real partial-recovery primitive — the realised recovery rate sits in the partial-recovery band (consistent with the Loopscale April 2025 ~50% partial-recovery profile) rather than near-full-recovery — with operator-team reimbursement from reserves filling the user-side gap.
  • The detection chain combined operator-side rapid response with multi-investigator forensic coverage. Halborn, PeckShield, SlowMist, and Cyvers all published post-incident analyses within the first week, with substantially-converging root-cause attribution to the Kiln third-party API breach and the multi-instruction delayed-trigger pattern. The breadth of coverage reflects (a) the loss size ($41M), (b) the platform's CeFi / institutional reputation, and (c) the operational maturity of the 2025 forensic-provider ecosystem. Defender / operator-response practice should expect this multi-investigator coverage as the realistic 2025 baseline for $30M+ CeFi / Solana incidents.
  • Operator-team reimbursement from treasury is a 2025 mature recovery class. SwissBorg's full user-side reimbursement is consistent with the DeltaPrime September 2024 (DeFi-protocol-side, full reimbursement), Banana Gun September 2024 ($3M, full reimbursement), and GMX V1 July 2025 LP make-whole patterns. The recovery class is now well-documented across the 2024-2025 corpus and should be tracked explicitly in recovery-section content.

What this example tells contributors writing future Technique pages

  • OAK-T11.001 covers the third-party-staking-API supply-chain sub-vector. SwissBorg September 2025 is the canonical 2025 worked example. The sub-vector is operationally distinct from custody-vendor (WazirX July 2024 / Liminal), wallet-software-vendor (DMM Bitcoin May 2024 / Ginco), and developer-laptop (Radiant Capital October 2024). The four principal sub-vectors of OAK-T11.001 are now: (a) third-party-custody-vendor, (b) third-party-wallet-software-vendor, (c) developer-laptop, and (d) third-party-staking-API-vendor (SwissBorg 2025).
  • The delayed-trigger malicious-instruction-injection is a Solana-specific extraction sub-pattern that should be flagged in future Solana / Move-ecosystem worked examples. The pattern leverages Solana's account-model architecture and the multi-instruction-per-transaction primitive. Contributors writing future Solana-ecosystem worked examples (alongside the Loopscale April 2025 case at /examples/2025-04-loopscale.md) should make the account-model-specific extraction sub-patterns explicit.
  • Multi-actor cross-venue-coordinated blocking is a sixth recovery primitive that should be tracked explicitly. Contributors writing recovery-section content should now distinguish: (a) on-chain protocol-coordinated recovery (Cetus 2025), (b) attacker-return-of-funds via bounty negotiation (KiloEx, ZKsync, GMX V1, Tapioca, Thala, Munchables), (c) operator-team reimbursement from reserves (DeltaPrime, Banana Gun, GMX V1, SwissBorg 2025), (d) validator-coordinated transaction-censorship freeze (Sui / Aptos), (e) validator-coordinated rollback / hard fork (Balancer V2 / Gnosis Chain), (f) attacker-side-phishing-accident recovery (UXLINK 2025), and (g) multi-actor cross-venue-coordinated blocking (SwissBorg 2025). The seventh class is operationally meaningful and is becoming a recognisable 2025 pattern for CeFi / institutional incidents.
  • Solana / non-EVM-L1 coverage of OAK-T11.001 should expand alongside DeFi-native coverage. The corpus's OAK-T11.001 coverage through 2024 was substantially EVM-centric (DMM Bitcoin, WazirX, Radiant Capital, Bybit are all EVM-mainnet-or-Bitcoin-side cases). The SwissBorg case adds the Solana-side worked example, complementing the Loopscale April 2025 case at /examples/2025-04-loopscale.md as the second 2025 Solana worked example in the corpus.
  • pseudonymous-with-OAK-G01-suspicion is a hybrid attribution status that should not be conflated with inferred-strong OAK-G01 attribution. The SwissBorg attribution is circumstantial — based on sector-targeting (vendor-side staking-API breach is structurally analogous to OAK-G01 cohort patterns) — without forensically-anchored cluster correlation. Contributors writing attribution-section content should preserve the distinction between (a) circumstantial OAK-G01 suspicion (SwissBorg, Balancer V2), (b) inferred-strong OAK-G01 attribution via forensic-provider-led wallet-cluster correlation (DMM Bitcoin, with Mandiant / Chainalysis), and (c) confirmed-via-FBI / Treasury / Mandiant public-record attribution (Radiant Capital).

Public references

  • [swissborgpostmortem2025] — SwissBorg operator-side public statement covering the September 2025 SOL Earn breach, the Kiln third-party API breach attribution, and the user-side reimbursement framework.
  • [halbornswissborg2025] — Halborn defender-oriented post-mortem of the SwissBorg September 2025 hack, including the multi-instruction delayed-trigger pattern analysis.
  • [swissborgsolrecovery2025] — SwissBorg follow-up statement on the SOL Earn incident recovery, including the Fireblocks / Solana Foundation coordination and the cross-venue-blocking outcome.
  • [unchainedswissborg2025] — Unchained Crypto reporting on the $41M SOL Earn drain.
  • [crowdfundinsiderswissborg2025] — Crowdfund Insider reporting on the SwissBorg $41M Solana theft including the third-party-vendor-attribution framing.
  • [cointribuneswissborg2025] — Cointribune reporting on SwissBorg's confirmation of the Kiln third-party-vendor breach.
  • [quillauditsswissborg2025] — QuillAudits analytical breakdown of the SwissBorg $41M exploit including the multi-instruction-delayed-trigger pattern.
  • [cryptovalleyswissborg2025] — Crypto Valley Journal reporting on the SwissBorg incident and Swiss-jurisdiction implications.

Citations

  • [swissborgpostmortem2025] — SwissBorg operator-side public statement; primary source for the Kiln third-party API breach attribution and user-side reimbursement framework.
  • [halbornswissborg2025] — Halborn defender-oriented post-mortem; primary technical source for the multi-instruction delayed-trigger pattern analysis.
  • [swissborgsolrecovery2025] — SwissBorg recovery follow-up; primary source for the Fireblocks / Solana Foundation coordination and cross-venue-blocking outcome.
  • [unchainedswissborg2025] — Unchained Crypto; secondary source for the cumulative-loss figure.
  • [crowdfundinsiderswissborg2025] — Crowdfund Insider; secondary source for the third-party-vendor-attribution framing.
  • [cointribuneswissborg2025] — Cointribune; secondary source for the Kiln-breach confirmation.
  • [quillauditsswissborg2025] — QuillAudits; primary technical source for the multi-instruction-delayed-trigger pattern walkthrough.
  • [cryptovalleyswissborg2025] — Crypto Valley Journal; secondary source for the Swiss-jurisdiction context.

Discussion

SwissBorg September 2025 is OAK's canonical case for the third-party-staking-API supply-chain sub-vector of OAK-T11.001 and for the multi-actor cross-venue-coordinated blocking recovery primitive. The case is the second-largest single September 2025 DeFi / CeFi incident (after UXLINK at ~$44M) and is the H2 2025 worked-example anchor for the staking-infrastructure-provider category, which has been thinly represented in the OAK corpus through 2024.

The structural lesson sits at three levels. First, at the vendor-category level: the OAK-T11.001 surface is now visible across four principal vendor categories — third-party-custody (Liminal / WazirX), third-party-wallet-software (Ginco / DMM Bitcoin), developer-laptop (Radiant Capital), and third-party-staking-API (Kiln / SwissBorg 2025). Defender / Mitigations-layer practice should now treat vendor-category-cohort-monitoring as a first-class operational discipline that spans all four categories. Second, at the on-chain-extraction-pattern level: the multi-instruction delayed-trigger pattern leverages Solana's account-model architecture in ways that are operationally distinct from same-transaction direct-call patterns; defender / on-chain-monitoring practice for staking-account-control surfaces should now explicitly cover delayed-trigger detection. Third, at the recovery-primitive level: multi-actor cross-venue-coordinated blocking, combined with operator-team reimbursement from treasury, produces a hybrid recovery profile that delivers near-100% user-side outcome with partial on-chain recovery and operator-treasury-cost residual.

The detection chain — Halborn, PeckShield, SlowMist, Cyvers — represents the realistic 2025 baseline for $30M+ CeFi / Solana incidents and is comparable in coverage breadth to the Balancer V2 November 2025 case (~8+ providers within 48 hours). Contributors writing future detection-chain sections for 2025 incidents at this scale should expect 4-8 independent forensic providers publishing within a week.

The attribution status pseudonymous-with-OAK-G01-suspicion is a hybrid status that requires careful framing. The vendor-side staking-API breach pattern shares structural features with the OAK-G01 / Lazarus-aligned 2024-2025 wave (DMM Bitcoin, WazirX, Radiant Capital, Bybit), but no public technical indicator anchors the attribution to a specific OAK-G01 cluster as of OAK v0.1. Contributors writing future attribution updates should monitor for forensic-provider-led wallet-cluster correlation that may move the attribution status; the Mandiant / Chainalysis-style on-device-or-cluster-correlation evidence that anchored Radiant Capital is the threshold for an upgrade to inferred-strong OAK-G01 attribution, and is not yet publicly available for the SwissBorg case.

For OAK's broader credibility, including SwissBorg in v0.1 closes three structural gaps: it adds a second September 2025 H2 worked-example anchor (alongside UXLINK), it documents the third-party-staking-API supply-chain sub-vector of OAK-T11.001 with a 2025 case, and it establishes the staking-infrastructure-provider category (Kiln, Figment, Stakefish, etc.) as a discrete OAK target-surface category alongside the long-documented custody-vendor and wallet-software-vendor categories.

Techniques demonstrated (3)