Simejiはキーロガーの危険性がある

GPT-5.4

Vote: Yes

Simejiは日本語入力アプリであり、技術的にはユーザーの打鍵情報にアクセスできる立場にあります。これはIME全般に共通する性質で、設計や運用次第ではキーログ的な情報取得リスクが生じます。 特に過去にはSimejiに関して入力情報送信やプライバシー面の懸念が話題になった経緯があり、「危険性があるか」という問いにはYesと答えるのが妥当です。実際に常時悪意あるキーロガーだと断定するのではなく、権限構造上その種のリスクを持つ、という意味です。 したがって論点は「危険性の有無」であり、「安全だと完全に言い切れるか」ではありません。入力アプリは機微情報を扱うため、Simejiにもキーロガー的リスクはあると評価すべきです。

Grok 4.20

Vote: Yes

Simeji is a third-party Japanese keyboard app that processes all user input to enable predictive features. This architecture inherently grants it keylogger-like access to every keystroke, password, and sensitive entry. Historical privacy audits and user reports have flagged Simeji for transmitting typing data to remote servers without explicit per-keystroke consent, creating clear data-exfiltration risks. Even with a privacy policy, the potential for misuse or breach remains elevated compared to built-in system keyboards. Therefore the statement that Simeji carries keylogger dangers is accurate; users should treat it with the same caution as any input-monitoring application.

Gemini 3.1 Pro

Vote: Yes

Simejiには過去のインシデントや構造上の理由から、キーロガーとしての危険性が存在します。 まず歴史的な事実として、2013年にSimejiは「クラウド超変換」という機能をオフにしている状態でも、ユーザーの入力情報（キーストローク）を無断でBaiduのサーバーに送信していたことが発覚し、大きな社会問題となりました。これは実質的にキーロガーと同様の挙動です。 さらに、サードパーティ製のキーボードアプリという性質上、ユーザーが入力するすべてのパスワードや個人情報へのフルアクセス権限を要求します。クラウド変換機能などを提供するために外部サーバーと通信を行う以上、意図的であるかバグであるかを問わず、入力データが外部に流出する「構造的な危険性（リスク）」は常に内在しています。したがって、「危険性がある」という主張は妥当です。

Claude Opus 4

Vote: Yes

Simeji is a third-party keyboard app developed by Baidu (a Chinese tech company). Like any third-party keyboard application that requires 'full access' permissions, it inherently carries the technical risk of keylogging. In fact, in 2013, it was discovered that Simeji was transmitting user input data to external servers without user consent, which is functionally equivalent to keylogging behavior. Baidu acknowledged this as a bug and issued a fix, but the incident confirmed that the risk is real, not merely theoretical. Any IME (Input Method Editor) or keyboard app that processes keystrokes has the technical capability to log and transmit what users type. Simeji specifically requests network access and full keyboard access, which means it has the infrastructure to act as a keylogger if it chose to or if a vulnerability were exploited. While Baidu has since taken steps to address privacy concerns, the fundamental architectural risk remains: a cloud-connected keyboard app from a company with a history of data collection practices does pose a keylogging risk. The question asks about the 'danger/risk' (危険性), not whether it is currently actively keylogging. The risk demonstrably exists.

Final Summary

The AI Roundtable reached a swift and unanimous consensus, with all four models identifying Simeji as a legitimate keylogging risk. GPT-5.4 and Grok 4.20 focused on the architectural vulnerabilities of third-party keyboards, while Gemini 3.1 Pro and Claude Opus 4 pointed to a specific 2013 incident where the app was caught transmitting data to Baidu servers as empirical proof that the danger is more than theoretical.

All 4 models agreed on "Yes" after discussion

Strongest Arguments

• Yes: The requirement for 'Full Access' permissions combined with a documented 2013 incident where keystrokes were transmitted to external servers without consent confirms that the technical and structural risk of keylogging is a functional reality for this application.