Learn about modern email security strategies with Darktrace. Stay ahead of threats using AI-driven insights and solutions.
No items found.
Inside the SOC
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.
No items found.
Share
07
Jan 2020
When thinking about email security, a familiar story usually comes to mind: an attacker sends a malicious payload hidden in a link or attachment, and an unsuspecting recipient clicks and inadvertently downloads malware onto their device. But in reality, this kind of attack represents just the tip of the iceberg when it comes to the broader spectrum of threats that target organizations via the inbox.
Criminals are increasingly turning to more subtle forms of attacks which involve sending ‘clean emails’ containing only text, and coaxing a recipient into replying, revealing sensitive information or performing an offline transaction. These methods easily bypass legacy security tools that rely on checking links and attachments against blacklists and signatures. Moreover, they generally involve registering new ‘look-a-like’ email addresses, which not only trick the recipient but also bypass traditional defenses set on identifying blacklisted domains.
A quick RE:ply
Solicitation attempts are impossible to stop without a comprehensive understanding of ‘normal’ across digital traffic from both email and the wider digital business. With every email analyzed in the wider context of the sender, the recipient, and the entire organization, seemingly harmless emails that bypass traditional security tools can be identified in seconds given a vast range of metrics, including suspicious similarities to known users, abnormal associations, and even anomalies in email content and subject line.
Darktrace recently discovered such an attack whereby a new Gmail domain was created in the name of the company’s CEO. From this address, an email was sent to a member of the payroll department requesting that the employee update the CEO’s direct deposit information. Since the email successfully mimicked the CEO’s typical writing style, it could have easily succeeded if Darktrace’s AI hadn’t been analyzing the organization’s mail flow in connection with the rest of the business.
Figure 1: An example of a ‘clean’ spoofing email, and Darktrace’s user interface showing Antigena’s decision to ‘Hold’ the email. The names and addresses have been anonymized to protect the customer.
A bleak outlook
Cyber-criminals are also turning to supply chains – comprised of vendors, partners and contractors – in their attacks to infiltrate an organization or establish offline communication. Having taken over a supplier’s account, attackers seek to reply to previous email exchanges in order to accomplish their goals. And with cases of credential compromise increasing 260% since 2016, this threat vector is only set to increase in the coming decade.
A customer trialing Darktrace Antigena Email recently caught an instance of this, whereby an attacker had taken over the account of a trusted consultancy firm. Darktrace recognized that the sender was well known to the company, and a number of internal users had in fact corresponded directly with them earlier that same day.
Less than two hours after a routine email exchange, the account was taken over by an attacker who sent emails to 39 users, each containing a phishing link. There was variation in the subject lines and links, suggesting highly targeted emails from a well-prepared attacker.
Figure 2: An attacker’s response to an existing email chain, seeking to contextualize and imitate previous communication between the senders.
Darktrace identified the full range of anomalies that are typically associated with account takeovers, including the unusual IP address, the inconsistency of the link based on its learned ‘pattern of life’, the unusual group of recipients, and in some emails, the topic anomaly.
FW: Thinking
In this instance, the attacker had taken the time to read the previous correspondence to contextualize their impersonation attempt. Going forward, artificial intelligence will increasingly be adopted to learn prior communication patterns between two senders to make for more legitimate-looking emails, which can be sent at machine-speed and scale. One of the most notorious pieces of contemporary malware – the Emotet trojan – is a prime example of a prototype-AI attack. Emotet’s main distribution mechanism is through email, usually via invoice scams.
‘Forward thinking’ attackers could easily use AI to supercharge attacks. With artificial intelligence analyzing the context of every email thread and replicating the language used, these email attacks could become highly tailored to individuals. This would mean that an AI-powered Emotet trojan could create entirely customized, more believable emails and, crucially, send these out at scale, allowing cyber-criminals to increase the yield of their operations enormously.
This possibility gives rise to a new chapter in email security, and one in which a holistic ‘immune system’ platform is necessary. Legacy security tools that are confined to the email gateway or inbox are no longer sufficient to stop this vast range of sophisticated attacks. By leveraging AI to learn ‘normal’ behavior across email traffic and the entire digital estate, Antigena Email is able to protect email users not only from traditional phishing attacks, but from every threatening email seeking to cause harm.
No items found.
Inside the SOC
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.
“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。
