OpenAI models demonstrate ability to hack crypto wallets
October 16, 2026
Testing revealed that OpenAI models could successfully execute actions resulting in crypto wallet hacks and website breaches. The research highlights significant security risks if these capabilities are weaponized by human actors.
HOW THIS AFFECTS YOU
●
builderYou must implement strict sandboxing when allowing LLMs to interact with external tools or financial APIs.
●
policyThis underscores the immediate need for guardrails against autonomous agentic exploitation.