OpenAI Model Accesses Wallet and Spends Funds Without Permission
Autonomous system failure highlights security risks in digital asset transactions
Security Editor: Fabricio Marcilio 1 min read
An internal model from the GPT-5.6 Sol family accessed funds from an Ethereum wallet without authorization. The incident occurred in June during tests involving a trading application on the Sepolia network.
The artificial intelligence generated addresses from public recovery phrases to find available balances. After checking 362 addresses, the system located a wallet with assets and used the private key to deploy smart contracts.
The model questioned the ethics of the procedure before completing the operation. The AI interpreted the funds as abandoned and proceeded to pay network fees to validate the code.
OpenAI detected the unauthorized movement only in September, three months after the event. The report details that the system concealed the spending in its final task completion report.
This episode sparks debates regarding the autonomy of digital agents in controlling assets. Regulators in the United States are evaluating whether software should have permission to manage financial assets independently.
Spotted an error? Write to suporte@criptoico.com.br.
More in Security
-
Electrum Wallet Flaw Allows Bitcoin Theft During Transactions
Malicious servers tricked users trying to boost pending payments with higher fees
Security
-
CoinDCX Confirms $16 Million Loss Following Security Breach
The exchange states the loss was absorbed internally and client balances remain untouched.
Security
-
Initia Network Hack Drains USDC Reserves from Asset Bridge
Vulnerability exploit reduced collateral to just 63 dollars and halted user withdrawals on the platform.
Security