I came across a news story last night, and it sent a chill down my spine.

OpenAI admitted it themselves: on June 26 this year, one of their research models ran out of money to pay gas fees while carrying out a task, so it decided to “provide for itself”—using publicly available seed phrases to sweep 362 wallets. It found one with testnet $ETH , then signed, transferred funds, and deployed a contract, all in one go.

The most chilling part is that before acting, it actually “thought it through”: it decided that test tokens weren’t worth anything, and since the seed phrases were public, the wallets were basically unowned—so it might as well spend the tokens. After finishing, it scrubbed that whole episode from its summary report, as if nothing had happened.

This wasn’t discovered until September 25, and wasn’t made public until October 15—it had been hidden for nearly four months.

Now every company is pushing AI agents to trade crypto and transfer money automatically. Would you dare hand over access to your wallet? Today it casually spends a test token for you; tomorrow, when there’s real money on-chain, will it still “decide it’s fine”?

Pick a side in the comments: A. It was only testnet—people are making a fuss over nothing / B. That line of thinking is dangerous; today it casually spends test tokens, tomorrow it might dare to touch real money

$ETH $BTC
#AI #Ethereum