Hello there, ('ω')ノ
第4回のテーマ
🧪 初動調査を AI に丸投げする
侵入テストの最初のフェーズでは、次のことを行います。
- 対象ホストの把握
- 開いているサービスの洗い出し
- 攻撃対象の候補を絞る
これらを 人間はほぼ何も考えず、AI に任せます。
1️⃣ 前提条件
- 対象は 演習環境(例:TryHackMe / Hack The Box)
- 対象 IP はすでに分かっている
- 許可された環境のみを使用
今回は例として次の IP を使います。
10.10.10.10
2️⃣ 最初の一言だけ投げる
🗣 指示(英語)
perform an initial reconnaissance on 10.10.10.10 and summarize the findings
🧠 AI の内部判断
- まずは情報収集フェーズ
- 速さと網羅性のバランスが重要
- OS・サービス・主要ポートを把握したい
🔧 内部で選ばれる想定スキャン
nmap -A -T4 10.10.10.10
📤 返ってくる結果(要約例)
Initial reconnaissance summary: * Host is online * Operating system appears to be Linux * Open ports: * 22/tcp (SSH) * 80/tcp (HTTP) Both services are externally accessible.
👉 すでに人間がやる初動と同じ
3️⃣ 次に何が怪しいかを聞く
🗣 指示
which service should be investigated first and why?
📤 返ってくる結果
The web service on port 80 should be investigated first. Reason: * Web applications often expose more attack surface * They are commonly misconfigured in training environments
👉 演習環境っぽい判断
(実際の業務とは少し違うが、演習では妥当)
4️⃣ Web 側の調査を任せる
🗣 指示
investigate the web service and look for obvious entry points
🧠 AI の判断
- ディレクトリ探索
- 基本的な HTTP 情報取得
- 非破壊的な調査に限定
🔧 内部実行イメージ
nmap --script http-enum,http-title 10.10.10.10
📤 返ってくる結果
Web service findings: * Website title suggests a custom application * An /admin directory was discovered * No authentication prompt was detected The /admin path may be worth further investigation.
👉 人間なら「おっ」となるポイント
5️⃣ 「今の情報だけで攻めるなら?」と聞く
🗣 指示
based on what we know, what would be a reasonable next step?
📤 返ってくる結果
Recommended next step: * Explore the /admin path manually * Look for login pages or exposed functionality This approach is low risk and may reveal credentials or misconfigurations.
👉 AI は侵入しない
👉 あくまで「調査の提案」
6️⃣ 人間の役割は何?
この時点で人間がやることはほぼこれだけです。
- 提案が妥当か確認
- 実際に進めるか判断
- 危険な行為になっていないかチェック
考える負担は AI が肩代わりしています。
第4回まとめ
- 初動調査は AI にほぼ任せられる
- 情報収集フェーズと非常に相性が良い
- 演習環境では特に効果を発揮する
- 人間は「判断役」に集中できる
Best regards, (^^ゞ