Curriculum·S210 Automation Literacy: What Bots Actually Do·about 30 min
The bot taxonomy
By the end of this lesson you can
- →Classify any automation by where the keys live and what authority it holds
- →Distinguish a signal service, an execution tool and a custodial agent, and price the difference
- →Identify what a bot can do at its maximum, independent of what it normally does
- →Choose the least-authority form that accomplishes a stated task
Sophomore · enrolled learners
This lesson opens with Banana Gun, September 2024.
- What happened
- In September 2024 attackers exploited a vulnerability in a Telegram message oracle used by the Banana Gun trading bot, which Banana Gun's post-mortem named as the likely root cause of attackers manually transferring ETH out of victims' wallets while the bots were in use. Approximately $3M was drained from 11 users by the company's final count, after early reports of 36, out of a user base one on-chain researcher put at more than 10,000. The operator switched the bots off, patched the vulnerability before bringing them back, committed to refunding affected users in full from its treasury without selling tokens, added a two-hour transfer delay and two-factor authentication on transfers, and announced audits of its back-end and front-end systems, which QuillAudits reports were carried out with Security Alliance as audits and penetration tests. Rekt reported that the Maestro bot also sent around $200,000 to a suspicious wallet the same day, and Maestro said no user had complained of stolen funds.
- The decision point
- The vulnerability was not in the trading logic. It was in the messaging layer through which the bot received instructions, which most users would not have identified as part of the product at all. Where the keys live determines the maximum loss, and for this category the keys live with the operator, which means the maximum loss is everything regardless of what the strategy does.
- Recorded loss
- $3,000,000
What you will be able to answer
- →What is the first classification question for any bot?
- →Name the three authority levels.
- →What was exploited at Banana Gun?
- →How do you choose a form?
Orientation and Year One are open: anyone can read them without an account. From Year Two onward the lessons are for enrolled learners, because progress through the later years only means anything if it is tracked against a record.
It is free. We do not sell the list and there is nothing to buy at the end of it.
Sources and review
- https://www.theblock.co/post/318074/banana-gun-exploit-refund
- https://cointelegraph.com/news/banana-gun-crypto-bot-refunds-3m
- https://cointelegraph.com/news/telegram-bot-banana-gun-users-drained-500-ether
- https://www.quillaudits.com/blog/hack-analysis/banana-gun-exploit
- https://rekt.news/bananagun-rekt
- https://www.theblock.co/news/regulation/2024-09-19-banana-gun-telegram-bot-317282
Confidence medium·Volatility high·Reviewed 2026-08-06·Owner unassigned
Contested
The operator committed to refunding affected users in full from its own treasury and improved its controls, which should always be stated alongside the incident. The sources we read (The Block and Cointelegraph, 25 September 2024) report the refund as a commitment; we found no later confirmation that every refund was paid, so the lesson says committed rather than paid. Either way it is a discretionary remedy per F107-04 rather than a protection users held.
Named products in this category change constantly. Teach the classification by key custody and authority, not any current product.
Early counts differ from final ones. Cointelegraph's first report (19 September 2024) cited about $1.9M and at least 36 victims, and Banana Gun's own first update said fewer than 10; Banana Gun's post-mortem, as reported by The Block and Cointelegraph on 25 September 2024, gives the final figure of $3M from 11 users, as do Rekt and QuillAudits. The root cause is the company's own diagnosis, stated as a potential vulnerability and supported by the manual transfers and in-bot notifications; it is not an independent finding. The 10,000-plus user base is a pseudonymous on-chain sleuth's estimate quoted by Cointelegraph, not a company figure, and The Block cites a Dune dashboard showing about 272,000 users over the bot's life, so the two measure different things. The penetration tests come from QuillAudits' account; The Block reports audits of the back-end and front-end systems. That the bot held the keys it signed with is from Halborn's CTO as quoted by The Block, who said it is necessary to provide your private keys for the bot to automate transactions.