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
Dan Fein
VP, Product
Share
10
Jun 2020
A report this week has revealed how the hackers-for-hire group known as ‘Dark Basin’ is targeting thousands of individuals with sophisticated and personalized email threats. The group has set up over 27,000 web pages to enable spear phishing attacks designed to harvest user credentials – most probably with the later intention of compromising the user’s account, eliciting sensitive information or wiring fraudulent payments. Among the thousands of individuals and organizations targeted are advocacy groups, journalists, elected officials, lawyers, and hedge funds.
The methods used by Dark Basin are extremely sophisticated – the emails were targeted to high-value individuals and aimed to gain their trust by falsifying known and trusted brands such as YouTube, DropBox or LinkedIn, or by posing as individual friends or colleagues. However, the reality is that thousands of businesses across the world are targeted by malicious emails crafted with this level of sophistication every single day. Darktrace regularly encounters email threats that leverage this same technique.
In fact, just last month Antigena Email neutralized an attack whereby a cyber-criminal spoofed the identity of a company’s CEO – writing in their exact style and tone – and sent out a heartfelt email to employees asking them to donate to a COVID-19 charity. The attacker had even taken the time to set up an authentic-looking web page with a donation form – all proceeds, of course, went directly into the threat-actor’s pockets.
This methodology of attack – whereby an attacker will impersonate a colleague, a boss, an IT department, or a trusted brand – has seen a significant rise this year. Prior to the outbreak of COVID-19 and the widespread adoption of remote working practices, around 20% of all malicious emails caught by Darktrace would have used some form of spoofing. Since March, we’ve seen that figure rise dramatically – 1 in 2 emails now contain some form of impersonation or spoofing. These sophisticated threats bypass the gateway on a daily basis, before being picked up and neutralized by Darktrace Cyber AI.
Figure 1: A graph showing the rise of impersonation attacks.
This trend reflects the overall success rate of this technique in the context of remote working. Last year, if you received an email from your colleague which seemed a little out of character, you might lean over your desk and ask them if they meant to send it. Today this is no longer possible. What is easier, making a phone call to check, or just clicking the link?
And it is not just individuals that these attackers are impersonating. A recent Darktrace blog gives examples where trusted presentation sites have been exploited to give a feeling of familiarity, and we’re seeing this extend to the full range of recognizable software brands we rely on for remote collaboration.
For example, we’ve seen many emails impersonating the Zoom platform, prompting the victim to accept an incoming ‘chat’ request from a colleague. The huge variety of different mechanisms that we all use to digitally communicate is playing into the hands of the criminals who suddenly find themselves with so many more methods to trick us.
Many of these malicious emails are now virtually indistinguishable from genuine communication — and there are no hard and fast rules for how employees can identify them. One email recently caught by Antigena Email attempted to coax the recipient into landing on a fake login page for the video conference application Zoom. The below illustrates how subtle the differences are between the counterfeit, and the genuine login page from the website.
Figure 2: Comparison of the counterfeit and genuine Zoom login pages.
Email filtering tools that compare emails against blacklists tend not to catch these more sophisticated and well-researched attacks. As discussed in a previous blog, many of these email threats rely on the creation of entirely new domains, which do not appear on these lists, and by default are let through.
An email security system that relies on this binary detection logic has a hard time differentiating between a legitimate email and a close copy, and no amount of employee training can guarantee complete immunity against these highly-convincing spoofing attacks. Furthermore, the ubiquity of information on social networks makes it easy for attackers to create believable emails.
However, technology powered by AI has been extremely successful in stopping these kind of advanced impersonation attacks by spotting subtle anomalies in emails that humans often miss. By understanding the human behind email communications, Antigena Email is the only email security technology that can ask whether it would be weird or unusual for a recipient to receive a given email, or visit a suspicious domain. Correlating insights around a sender’s login location, the extent of prior communication, the rarity and location of links, and over 750 other metrics, the technology detects the subtle hallmarks of an email attack that other tools miss.
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。
