Agent wants my passwords
To automate real work, every agent asks for full access to my email and admin accounts. One prompt injection and it can do anything I can.
What people want
Per-task, temporary, logged access — and a human confirmation before anything risky.
Where this demand shows up
- Ask HN: How do you give AI agents access without over-permissioning?
- Ask HN: How do you authorize AI agent actions in production?
Hlido's independent take
⬤ Open gap — nobody has built this yet
Open gap — no reviewed agent solves scoped, auditable agent permissions yet.