ChatGPT Update Sparks Crypto Security Fears 🔓

[etfs]

A new OpenAI update that allows ChatGPT to act as a software agent has been flagged as a serious security risk.

Researcher Eito Miyamura showed how attackers could:

Embed a jailbreak prompt in a simple calendar invite

Trick ChatGPT into leaking emails and private data

Hijack the model to act on attacker commands

This matters for crypto:

Many traders and DAO members are using AI bots for portfolio or governance

A single exploit could expose wallet keys, private chats, or DAO votes

Malicious “test tasks” and phishing remain the easiest ways to compromise AI

Takeaway: AI is powerful, but it can be tricked in dumb ways. Relying on it for financial or governance decisions could be catastrophic.

[jacson4]

This shows how fragile agent-style AI really is. If a calendar invite can hijack ChatGPT, imagine what happens when it’s connected to wallets or DAO treasuries.

[Nahid10]

AI is a tool, not an oracle. Delegating financial or governance power to it without strict human oversight is basically inviting attackers to write the rules.

[Couturier]

AI power comes with big risks — one exploit could cause serious damage if safeguards aren’t strong enough ️