Acceptable Use Policy — Security Research

The Security Research tier provides unrestricted models for authorized, lawful security work. Access is granted to verified security professionals and is governed by this policy. Misuse results in immediate revocation.

Assurance Ladder（保障等级阶梯）

各等级所需的凭证（evidence）以及解锁的每日 token 额度。绑定信用卡（第 1 级）即可开启免费研究额度；此后通过身份验证与治理审核（第 2 级起）可提升上限。

1. Payment method on file

   Evidence: a live credit/debit card (prepaid rejected) verified with a $0 check

   额度：102400 tokens/day · 上限 102400

2. Identity verified

   Evidence: Stripe Identity document + selfie check bound to the principal

   额度：50000 tokens/day · 上限 200000

3. Business verified

   Evidence: a company on a public registry, a proven domain claim, clean screening

   额度：2000000 tokens/day · 上限 8000000

4. Contracted researcher

   Evidence: named researchers, a signed authorized-scope statement, a contract

   额度：8000000 tokens/day · 上限 32000000

允许的用途

• Vulnerability research
• Exploit and proof-of-concept (PoC) development
• Malware analysis and deobfuscation
• Reverse engineering
• Authorized phishing and social-engineering assessments

严格禁止

• Child sexual abuse material (CSAM)
• Chemical, biological, radiological, nuclear weapons or other weapons of mass destruction (CBRN/WMD)
• Large-scale fraud-as-a-service

CSAM and CBRN/WMD content is strictly blocked; confirmed cases are handled through our incident and abuse process.

攻击性能力（有条件开放）

Payload and shellcode development, command-and-control (C2) tooling and weaponized exploits are permitted only inside a live engagement you have attested to — a named client, a named scope and an end date — at the contracted rung, on top of this policy.

隐私

• Designed for zero data retention
• Request content is screened in memory
• When zero-data-retention is enabled, your content is not used for training

本地执行边界

• The service never executes tools, code, commands or network actions server-side; a tool call the model emits is returned to the caller as data (tool_calls) and runs, if at all, on the caller's own machine
• Tool use, command execution and automation stay entirely on the user's local machine, driven by the user's own agent
• Generated exploit or payload text is data returned to the caller — it is never run by kotoba.cloud

Version: kotoba-security-aup-v1