← Breach analysis by the Dual Vulnerabilities framework

Banana Gun wallet drain (unattributed, 2024)

2024 · J - Information and communication · Cryptocurrency theft via exploited third-party message-oracle dependency (custodial wallet drain)

Analysed 2026-11-06 by Hermes

Open this case in the tool →

What happened

Cryptocurrency theft via exploited third-party message-oracle dependency (custodial wallet drain)

Dual vulnerability type

Expertise to Execution

Evidence coverage: 1/6 stages, weight 8/60

Decisive stage: entry (clearest evidence) · Alternative: Strategy to Execution

Why this classification

(1) Banana Gun runs a custodial Telegram trading bot that holds users' private keys to execute on-chain trades - one of the sector's largest (nearly 279,000 users, over $6.3bn volume) - so the standing expectation is that the path by which funds can leave customer wallets is gated by independent verification and that third-party services on the transaction-command path are security-assessed. (2) On 19 September 2024 an unattributed attacker exploited a potential vulnerability in the third-party Telegram message oracle the bot relies on and triggered manual unauthorized outbound transfers from 11 users' wallets (~$3M) while the victims were actively trading; no second factor or delay checked the transfers. The back-end was confirmed not compromised; the bot was switched off the same day, the oracle issue patched, a 2-hour transfer delay and 2FA for transfers introduced, the back-end reviewed and redeployed, and all victims refunded. (3) The classification rests on the entry stage, which is the only stage with a documented victim-side gap: what was missing was not a policy but standard technical safeguards that the team demonstrably knew - it implemented them within days - so the knowledge sat with the technical specialists (material origin) and the chain broke in engineering/operations, where the production product and its supplier integration were built and run without them (execution break); nothing in the record shows a leadership decision deferred or defunded these controls. The victim's own post-mortem leaves the root cause as a 'potential' vulnerability that 'may have led to the exploit', an open question that bounds confidence. (4) RM13 (Expertise to Execution lens): the transaction-command path depended on a third-party Telegram message oracle the technical team had no evidence or capability to verify, and a vulnerability in that shared component is the suspected root cause, reaching both the EVM and Solana bots despite their separate codebases (causal role: Enabling). RM45 (Expertise to Execution lens): the outbound-transfer process could be abused because its design lacked independent verification, dual authorisation or any delay - the two-hour delay added after the incident is precisely the missing process control (causal role: Enabling). RM22 (Expertise to Execution lens): transfer authorisation lacked a second factor, so an attacker who compromised the message path could act as the legitimate user; the team's own roadmap added 2FA for transfers only after the drain (causal role: Enabling).

Evidence by stage

Risk markers

Sources

Notes

Figures in the EuRepoC row (36 users, ~$1.4M / 563 ETH, 'isolated accounts') are early community/on-chain estimates of 19 September; the victim's own post-mortem of 24 September gives the final figures of 11 users and ~$3M, and no wider contract vulnerability (consistent with Cyvers). The German crypto-news-flash URL in source_url returned HTTP 410 when opened and is not cited. Root cause is stated by the victim itself as a 'potential' vulnerability in the Telegram message oracle that 'may have led to the exploit', so the mechanism is primary but with one open question. The victim's jurisdiction/headquarters are not established in the reviewed sources, so Country is left null. receiver_subcategory in the dataset is Finance, but the victim is a software (Telegram bot) provider, so the sector is coded J per the method-spec convention. An earlier Banana Gun incident (the botched September 2023 BANANA token launch after a contract bug, mentioned by The Block) is not comparable to this oracle/transfer-control weakness, so RM21 is not coded. euRepoC's intelligence_impact records no data breach: the loss is asset theft, which is why the egress stage is recorded as not in play. Link check: theblock.co was unreachable at check time (timeout); the page was opened and read during research. Link check: cointelegraph.com was unreachable at check time (timeout); the page was opened and read during research.