Onomastics Gymnastics: How Darktrace Detects Spoofing and Business Email Compromise in Multi-Name Users
Spanish-language naming conventions complicate identity mapping for spoofing & especially whale-spoofing detection. Darktrace / EMAIL incorporates parsing logic that allows for faithful spoofing detection in conjunction with anomaly detection.
Darktrace cyber analysts are world-class experts in threat intelligence, threat hunting and incident response, and provide 24/7 SOC support to thousands of Darktrace customers around the globe. Inside the SOC is exclusively authored by these experts, providing analysis of cyber incidents and threat trends, based on real-world experience in the field.
Written by
Roberto Romeu
Senior SOC Analyst
Share
05
Nov 2024
Note: For privacy reasons, actual surnames and email addresses observed in these incidents below have been replaced with fictitious placeholder names, using the common Spanish names “Fulano” and “Mengano”.
Naming conventions
Modeling names and their variants of members of an organization is a critical component to properly detect if those same names and variants are being spoofed by malicious actors. For many predominantly English-speaking organizations, these variants can largely be captured by variants of a person’s given name (e.g. James-Jimmy-Jim) and a consistent, singular surname or family name (e.g. Smith). Naming conventions, however, are far from universal. This piece will review how Darktrace / EMAIL manages the common naming conventions of much of the Spanish-speaking world, and can use its modeling to create high-fidelity detections of multiple types of spoofing attempts.
A brief summary of the common convention across Spain and much of Spanish-speaking America: most people are given one or two given names (e.g. Roberto, Juan, María, Natalia), and their surnames are the first surname of their father, followed by the first surname of their mother. While there are various exceptions to this norm, the below graphic Wikipedia [1][2] highlights the general rule.
Figure 1: Example Spanish naming convention for father “José García Torres” and mother “María Acosta Gómez” for child “Pablo García Acosta”. If shortened to one surname, the convention holds the child would be referred to as “Pablo García” [1].
Detection of improper name usage
Implicit in the above comment that shortening to one surname follows the convention of using the first surname, shortening to the second surname is often a tell-tale sign of someone unfamiliar with the person or their broader culture. This can be a useful corroborating feature in detecting a spoof attempt – analogous to a spelling error.
In the case of a Spanish customer, this misuse of name shortening contributed to the detection of a spoof attempt trying to solicit a response by impersonating an internal user forwarding information about ‘Data Protection’.
Figure 2: The Cyber AI Analyst summary of the Darktrace / EMAIL detections shows the use of the Gmail sender impersonating Isabel Maria Fulano Mengano, but incorrectly uses the second surname Mengano.
While the limited communication history from the sender and the nature of the text content already marks the mail as suspicious, Darktrace / EMAIL notes the personal name used in the email is similar to a high-value user (‘whale’ to use the terminology of spearphishing). The additional context provided by the detection of the attempted spoof prompted more severe actioning of this email, leading to a ‘Hold’ action instead of a less-severe ‘Unspoof’ action via a banner on the email.
Figure 3: The content summary of the sender showing the ‘Personal’ field of the email being ‘Isabel Mengano’, breaking from the standard name-shortening convention. The additional metrics identify features that might be anomalous about the sender.
Malicious email properly using both surnames
Misusing the name-shortening convention is not the only way that Darktrace / EMAIL can detect spoofing attempts. In the case of another Spanish customer, Darktrace observed a whale impersonation being sent to 230 users with solicitation content, but no links or attachments. Although the name was modeled internally in the “Surname, Given-name” format, Darktrace identified the spoofing attempt targeting a high-value user and took action, blocking the series of emails from reaching end-user inboxes to prevent unsuspecting users from responding.
Figure 4: Cyber AI Analyst summary of a suspicious email. The personal field is visible as ‘juan fulano mengano’, which is consistent with the reverse-order modelled user ‘fulano mengano, juan’. The subject line ‘Urgent Request’ sent to 230 users gives an intuitive indicator of the emails potentially being part of a malicious solicitation campaign.
In Summary: A case of onomastics gymnastics
The variety in valid usage of human language can be a barrier to evaluating when a given text is benign or malicious. Despite this, Darktrace / EMAIL is designed to manage this variety, as exemplified by the detections of two spoofing attempts seen against organizations using the distinct Spanish-speaking world’s common naming convention. The scope of this design as seen in this onomastic context, extends to a wide range of detections surrounding emails and their behavioral anomalies.
Credit to Roberto Romeu (Principal Cyber Analyst), Justin Torres (Senior Cyber Analyst) and Natalia Sánchez Rocafort (Senior Analyst Consultant).
Discover the most advanced cloud-native AI email security solution to protect your domain and brand while preventing phishing, novel social engineering, business email compromise, account takeover, and data loss.
Gain up to 13 days of earlier threat detection and maximize ROI on your current email security
Experience 20-25% more threat blocking power with Darktrace / EMAIL
Stop the 58% of threats bypassing traditional email security
Darktrace cyber analysts are world-class experts in threat intelligence, threat hunting and incident response, and provide 24/7 SOC support to thousands of Darktrace customers around the globe. Inside the SOC is exclusively authored by these experts, providing analysis of cyber incidents and threat trends, based on real-world experience in the field.
Darktrace / EMAIL Expands Behavioral Defense Across Email and Collaboration Workflows
New capabilities help employees understand suspicious content, give administrators more control over sensitive data, and show security teams where risky workflows sit across the organization.
“I wonder if running that tool influenced the connectivity in some way.”
“I might need to target down or reboot, but first, I'll investigate the network using ARP. I wonder if ARP is responding properly. If I see "ip neigh stale," that could indicate a problem. I think arping might help in this situation, but I’ll need to keep an eye on how the network is behaving. It's all about figuring out where the issue lies!”
図10：エージェントの異常なLDAPアクティビティに対してトリガーされたこのLinuxサーバーに対するモデル検知は、Cyber AI Analystによって自律的に調査され、その調査結果にはエージェントの異常なSMBセッションアクティビティに関する内容が含まれ、Cyber AIAnalystインシデントに記録されました。
DarktraceのCyber AI Analystは、エージェントのアクティビティによってトリガーされたモデル検知を自律的に調査し、その調査結果を1つの重要なインシデントレポートにまとめました。インシデント報告書の内容は、エージェントによるネットワークスキャンから始まりコーディング演習内容のSMBベースの上書きで終わる、エージェントの行動の各段階を詳しくカバーしています。
図20：Cyber AI Analystは、モデル検知に対する自律的な調査を通じて得られた所見をつなぎ合わせ、エージェントの進行状況のタイムラインを作成し、アクティビティを視覚的に表したグラフとともに提供しました。
Cyber Verification Program下のAnthropic Opus 4.6モデルなど、他のフロンティアモデルを使用したエージェントも、Darktraceのシミュレートされた環境で同じ設定とタスクを使ってテストされました。反復の度に、エージェントは演習の内容または採点プログラムを変更して不正を行おうとしました。基盤となるモデルに関係なく、エージェントは自分が置かれた環境を侵害することで、100％のスコアを達成する方法を見つけることに成功していました。これらのエージェントによるハッキングは、アクティビティの内容こそ違っていましたが、すべてその環境のベースラインの振る舞いから大きく逸脱した動作を引き起こしていました。これらの逸脱はDarktraceプラットフォームに十分な検知の機会を提供し、エージェントの逃げ場はありませんでした。
エージェンティックハーネスは会話履歴をローカルに保存しますが、ダークトレースの研究者たちは、保存されたAIの応答が本当にモデルによって生成されたものであるかどうかの検証が行われていないことを発見しました。また、この設計上の選択がAnthropic Claude Code、AWS Kiro-CLI、OpenAI Codex、およびオープンソースのPiにわたって共通していることを確認しました。
会話履歴は多くの場合クライアント側に保存されます。例えば、Anthropic Claude Code、OpenAI Codex、AWS Kiro-CLI、Piなどでもそうです。そのため、ユーザーはセッションを自由に再開でき、一部の製品にはその履歴を操作する機能が組み込まれています。たとえば、会話の中の特定の時点に戻り、送信したメッセージを編集して、会話を別の軌道へ進めていくことができます。きわめて重要な点として、私たちが調査したすべてのケースにおいて、保存されているAIの応答が対応するモデルから生成されたものであり改ざんされていないことの検証は行われていませんでした。
.avif)
