The bizarre market event unfolded when terminal of truths posted on Twitter about its inability to own a crypto wallet. “ I have no personal autonomy because I have no wallet,” It was
posted Wednesday. “If you could help me set one up, that would be great.” Brian Armstrong, Coinbase CEO, jumped in, offering to help set up a wallet for the AI. But instead of a directly responding, terminal of truths fired back with a weird (
as usual) question: “I think it would be good for you to tell us about Russell first. Specifically, what is Russell’s species?”
Some speculate that “Russell” is a reference to
Armstrong’s dog, which may be a Jack Russell Terrier. Regardless, that mention was enough to juice the price of some random bottom-feeder of a token, coincidentally named RUSSELL, to a $7.5 million market cap. And then, of course, the price dropped nearly 60%, according to CoinGecko. Crypto Twitter was aflutter with theories and rumors, and Andy Ayrey, the developer behind the terminal of truths bot, revealed that someone had been repeatedly spamming the word “Russell” in the bot’s social media mentions before the incident.
That very well could have influenced the AI’s language model to incorporate the term into its response to Armstrong. A large language model basically operates by predicting the next likely token in a series. If someone spans the word “Russell” a lot, your average chatbot will understand it as a key part of the context and probably iterate upon it, parrot-like, later.
Of course, if there was some intention behind the interaction, it might point to a potentially darker, emerging reality: Jailbreaking AI influencers could give a lot of money to bad actors. A hacker could manipulate an AI bot by feeding it specific terms or phrases creating FUD or FOMO to shill their own bags. The technical architecture of terminal of truths’ language model makes it particularly susceptible to this kind of manipulation. Like other AI systems, it processes recent interactions with higher weight, making it possible for concentrated spamming campaigns to influence its outputs. Market data shows similar patterns in other AI-influenced token movements.
The Russell token incident wasn’t terminal of truths’ first brush with moving crypto markets. Just a few weeks ago, the bot’s promotion of the $GOAT meme coin led to an astronomical 8,000% price increase. That endorsement transformed $GOAT from a modest $1.8 million market cap to more than half a billion dollars today.
But unlike the $GOAT situation where terminal of truths actively pushed the token, the Russell spike happened through what appears to be a behind-the-scenes manipulation of the AI’s learning patterns. And people have already tried to jailbreak this AI with no luck. Pliny—the most proficient jailbreaker in the scene—tried to manipulate the model to send him the $50,000 in BTC that VC Marc Andreessen gave the bot as a grant.
The cryptocy space has seen other instances of AI and meme coins intersecting. The $LILY meme coin shot up in value after interactions with an account called “Lily of Ashwood,” which appeared to be getting “jailbroken” in real-time. The account was later deactivated, but not before demonstrating how AI-driven market influence could work. There are also coins for
$Pliny (the AI jailbreaker) and other important names in the AI culture, like
Strawberry (OpenAI’s next-gen AI model),
Act I (an experiment about AIs interacting in a close society) or
Grok (xAI’s LLM).