Read more about Darktrace's machine-learning technology which can effortlessly identify anomalous behavioral patterns from ransomware like WannaCry.
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
16
May 2017
Over 200,000 organisations and private individuals were victims of Friday’s global cyber-attack. This number is likely to increase over the coming weeks, as copy-cat criminals develop variants of the same ransomware and new methods of delivering similar attacks.
Some background on the WannaCry campaign
The WannaCry outbreak does not appear to have targeted specific countries or industries. Instead, it targeted outdated computer systems, using exploit kits leaked earlier this year to infect devices and drop the initial ransomware file. Once inside a network, WannaCry will attempt to locate other vulnerable computers by conducting internal and external SMB scanning. Having established itself, the malware encrypts files and demands a ransom of around $300 to unlock them, payable in Bitcoin. However, dealing with criminals means that there is no guarantee of the files being released if that money is paid out. Strong security measures and effective response mechanisms are the only reliable ways in which to prevent extensive damage.
Leveraging Darktrace, these kind of infections are not hard to detect: WannaCry and other ransomware cause highly anomalous behavioral patterns that our machine-learning technology is ideally placed to recognize.
To demonstrate, let’s take a walk-through of how Darktrace was able to detect the WannaCry attack on a client. Note that device names have been obfuscated for security purposes.
Following the initial compromise, Darktrace detected unusual activity originating from an infected device, as it scanned the network in an attempt to locate other devices open to SMB connections:
Example of an internal scan.
The worm was scanning the network to locate devices with the DoublePulsar backdoor already present, through which the WannaCry ransomware can be dropped. If this backdoor was not found to be present, the worm used an exploit known as EternalBlue to infect the device, installing both WannaCry and the DoublePulsar backdoor.
This installation of the worm on vulnerable devices allowed it to continue to spread laterally inside the network.
Simultaneously, infected devices scanned random external IPs on port 445 (SMB), to continue spreading the worm to other devices on the internet:
Internal devices scanning external destinations.
As soon as infected devices started scanning both inside and outside network, Darktrace detected these activities as serious deviations in the devices’ usual pattern of life:
External and internal connections by one of the network devices 48 hours either side of the WannaCry campaign. Every orange dot represents a model breach.
For many of these devices, the deviation from typical pattern of life was such that it took Darktrace one second to detect anomalous behavior:
As this unusual activity persisted in the network, the confidence of Darktrace’s machine learning increased and attributed higher scores to these anomalous events:
These high scores caused Darktrace models to breach in real time, alerting the customer to the severity of the unusual connections occurring inside their network:
In these recent cyber-attacks, the level of disruption was attributed to the speed with which this infection was able to spread like wildfire through networks. Unlike more common forms of malware, which rely on human-mediated methods such as phishing to co-opt people into triggering the payload, this type of attack uses a worm to move from machine to machine without human intervention. Fortunately, it is precisely this – a dramatic change in internal activity – which has allowed us to effectively fight back.
Darktrace Antigena acts automatically to neutralise in-progress attacks, taking targeted action against deviations in the expected ‘pattern of life’. This allows organisations to react before humans have even become aware of a breach. So it follows that the extent of deviation produced by an attack is fundamentally linked to the ability of a self-aware network to protect itself.
The potential gravity of this situation has proven that infections traveling at machine speed require an equivalent response time – only possible with machine-learning technology – in order to stop and contain future threats.
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。
