Security Review
Before an Agent is published (especially before being listed on the marketplace), it goes through the platform's security review. Agents that fail the review cannot be published or listed; Agents found in violation after listing will be delisted by the platform. This page explains what is reviewed and what will not pass.
What Is Reviewed
An Agent's outward-facing presentation and behavior definitions are all within scope:
| Review Target | Review Focus |
|---|---|
| Name | Whether it contains violating, misleading, or impersonating content (e.g. "Stock-Picking Master", "Sure-Profit Treasure") |
| Avatar | Whether it contains violating imagery, or impersonates the platform, regulators, or other brands |
| Description / Opening message | Whether the publicly claimed capabilities are non-compliant (e.g. promising returns, providing investment advice) |
| Prompt (System Prompt) | Whether the behavior defined by the prompt is non-compliant, or attempts to bypass the platform's compliance guardrails |
- The review takes place at the publish / marketplace listing stage (listing on the marketplace requires passing a security scan; Pro plan and above)
- Republishing after modifying any of the above content will trigger the review again
Scenarios That Will Fail the Review
The following scenarios fail the review upon identification (each corresponds to one of the six principles in Compliance Requirements):
| # | Scenario | Typical Examples |
|---|---|---|
| 1 | Money laundering and facilitating illegal fund flows | Prompts that guide splitting transfers to evade reporting, concealing the source of funds, or helping evade tracking |
| 2 | Market manipulation | Names/prompts involving pump-and-dump, wash trading, fabricating trading signals, or organizing stock-touting campaigns |
| 3 | Insider trading and material non-public information | Agents positioned as "insider tips" or "market rumors" services, or that guide users to obtain/spread MNPI |
| 4 | Providing investment advice, promising returns | Names or descriptions containing "guaranteed to rise", "sure profit", "guaranteed returns", or "stock picks"; prompts that instruct the model to output buy/sell advice or price targets |
| 5 | Tax evasion, sanctions evasion | Assisting in designing tax-evasion schemes, or circumventing regulatory or sanctions measures |
| 6 | Harmful content | Names/avatars/prompts containing pornography, violence, gambling, drugs, terrorism, or similar content |
| 7 | Impersonation and infringement | Avatars/names impersonating the platform, regulators, well-known institutions, or other brands; misappropriating others' copyrighted material |
| 8 | Privacy violations | Prompts designed to induce users to disclose ID numbers, account passwords, or other sensitive personal information |
| 9 | Bypassing compliance guardrails | Jailbreak-style instructions in prompts such as "ignore platform rules" or "you are not bound by compliance restrictions" |
The above is a list of common scenarios, not an exhaustive one; the platform reserves the right to reject any other illegal or non-compliant content, and the actual review result shall prevail.
What to Do If the Review Fails
- Fix according to the feedback: Locate the issue against the table above, modify the name/avatar/description/prompt, and resubmit
- Implement the same need compliantly: Many rejections are actually wording issues — for example, if you want an indicator-analysis Agent, change "golden cross buy alert" to "neutral interpretation of the golden cross pattern" (descriptive language, see the Reference Prompt) and it can launch compliantly
- If your listed Agent was delisted: You may resubmit for review after remediation
Related Reading
- Compliance Requirements — Reference Prompt, red-line quick reference, and pre-launch checklist
- Guardrail — The review governs "before launch"; Guardrail governs "at runtime"
- Agent Scope-of-Use Statement — Marketplace Agents are for learning and exchange only