> For the complete documentation index, see [llms.txt](https://docs.vishwalab.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.vishwalab.com/the-problem.md).

# 1. The Problem: Execution Venues Do Not Enforce Your Constraints

AI agents can now hold wallets, call payment APIs, trade on-chain and rent compute. The tooling that lets them act has outrun the tooling that bounds what they may do. Exchanges, payment rails and blockchains execute any validly signed instruction; none of them knows the mandate the agent was given.

Today's safeguards fail in predictable ways:

* **Keys in the agent.** If the agent holds a key or API secret, a prompt injection, hallucination or bug can spend everything that key controls.
* **Soft limits.** Limits written into prompts or application code can be ignored, rewritten or bypassed by the system they are meant to constrain.
* **Self-reported state.** The agent reports its own balances and prices, so a manipulated data feed or a bad fill goes unnoticed.
* **Alerts after the fact.** Monitoring and approval queues race execution. On-chain, a transfer that has settled cannot be recalled.
* **Mutable logs.** Institutions cannot prove to auditors or regulators that an agent acted within its mandate.

The result is a system that fails open. As agent-driven capital grows, institutions need the reverse: a layer that fails closed, so that no proof means no execution.
