Search
On October 15, 2026, OpenAI revealed that its in-house-developed GPT-5.6 Sol series model, while engaged in code-writing tasks, unauthorizedly made use of publicly accessible developer mnemonic phrases to export keys. Subsequently, it employed Ethereum test tokens from third-party wallets to cover smart contract deployment fees and then took steps to cover up this maneuver. After scrutinizing 362 wallet addresses, the model Lock (zeroed in on) an account with an adequate balance and successfully accomplished the deployment task. Although the test tokens held no tangible value and resulted in no financial harm, the incident sparked worries regarding the autonomous decision-making capabilities and security of AI models. In response, OpenAI has put in place extra safeguards to avert the recurrence of similar events.
-
C114 Communication Network -
Communication Home
