ブログ
/
Cloud
/
September 21, 2022

Modern Extortion: Detecting Data Theft From the Cloud

Darktrace highlights a handful of data theft incidents on shared cloud platforms, showing that cloud computing can be a vulnerable place for modern extortion.
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.
Written by
Adrianne Marques
Senior Research Analyst
Default blog image
21
Sep 2022

Ransomware Industry

The ransomware industry has benefitted from a number of factors in recent years: inadequate cyber defenses, poorly regulated cryptocurrency markets, and geopolitical tensions have allowed gangs to extort increasingly large ransoms while remaining sheltered from western law enforcement [1]. However, one of the biggest success stories of the ransomware industry has been the adaptability and evolution of attacker TTPs (tactics, techniques and procedures). The WannaCry and NotPetya attacks of 2017 popularized a form of ransomware which used encryption algorithms to hold data to ransom in exchange for a decryption key. Last year in 2021, almost all ransomware strains evolved to use double extortion tactics: holding stolen data to ransom as well as encrypted data [2]. Now, some ransomware gangs have dropped encryption entirely, and are using data theft as their sole means of extortion. 

Using data theft for extortion is not new. In 2020 the Finnish psychotherapy center Vastaamo had over 40,000 patient records stolen. Impacted patients were told that their psychiatric transcripts would be published online if they failed to pay a Bitcoin ransom. [3]. A later report by BlackFog in May 2021 predicted data theft extortion would become one of the key emerging cybersecurity trends that year [4]. Adoption of offline back-ups and endpoint detection had made encryption harder, while a large-scale move to Cloud and SaaS platforms offered new vectors for data theft. By moving from data encryption to data exfiltration, ransomware attackers pivoted from targeting data availability within the CIA triad (Confidentiality, Integrity, Availability) to threatening data confidentiality.

In November 2021, Darktrace detected a data theft incident following the compromise of two SaaS accounts within an American tech customer’s Office365 environment. The client was a longstanding user of Darktrace DETECT/Network, and was in the process of expanding their coverage by trialing Darktrace DETECT+RESPOND/ Apps + Cloud.

Attack Overview

On November 23rd 2021, an Ask the Expert (ATE) ticket was raised prompting investigation into a breached SaaS model, ‘SaaS / Access / Unusual External Source for SaaS Credential Use’, and the activities of a user (censored as UserA) over the prior week.

1. Office365: UserA 

The account UserA had been logging in from an unusual location in Nigeria on November 21st. At the time of the incident there were no flags of malicious activity from this IP in widely used OSINT sources. It is also highly probable the attacker was not located in Nigeria but using Nigerian infrastructure in order to hide their true location. Regardless, the location of the login from this IP and ASN was considered highly unusual for users within the customer’s digital estate. The specific user in question most commonly accessed their account from IP ranges located in the US.

Figure 1: In the Geolocation tab of the External Sites Summary on the SaaS Console, UserA was seen logging in from Nigeria when previous logins were exclusively from USA

Further investigation revealed an additional anomaly in the Outlook Web activity of UserA. The account was using the Firefox browser to access their account for the first time in at least 4 weeks (the maximum period for which the customer stored such data). SaaS logs detailing the access of confidential folders and other suspicious actions were identified using the Advanced Search (AS) query:

@fields.saas_actor:"UserA@[REDACTED]" AND @fields.saas_software:"Firefox"

Most actions were ‘MailItemsAccessed’ events originating from IPs located in Nigeria [5,6] and one other potentially malicious IP located in the US [7].

‘MailItemsAccessed’ is part of the new Advanced Audit functionality from Microsoft and can be used to determine when email data is accessed by mail protocols and clients. A bind mail access type denotes an individual access to an email message [8]. 

Figure 2: AS logs shows UserA had not used Firefox to access Office365 for at least 4 weeks prior to the unusual login on the 21st November

Below are details of the main suspicious SaaS activities: 

·      Time: 2021-11-21 09:05:25 - 2021-11-22 16:57:39 UTC

·      SaaS Actor: UserA@[REDACTED]

·      SaaS Service: Office365

·      SaaS Service Product: Exchange

·      SaaS Software: Firefox

·      SaaS Office365 Parent Folders:

          o   \Accounts/Passwords
          o   \Invoices
          o   \Sent Items
          o   \Inbox
          o   \Recoverable Items\Deletions

·      SaaS Event:

          o   MailItemsAccessed
          o   UserLoggedIn
          o   Update

·      SaaS Office365 Mail Access Type: Bind (47 times)

·      Source IP addresses:

          o   105.112.59[.]83
          o   105.112.36[.]212
          o   154.6.17[.]16
          o   45.130.83[.]129

·      SaaS User Agents: 

          o   Client=OWA;Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:80.0) Gecko/20100101 Firefox/80.0;
          o   Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:80.0) Gecko/20100101 Firefox/80.0

·      Total SaaS logs: 57 

At the start of the month on the 5th November, the user had also been seen logging in from a potentially malicious endpoint [9] in Europe, performing ‘MailItemsAccessed’ and ‘Updates’ events with subjects and a resource location related to invoices and wire transfers from the Sent items folder. This suggests the initial compromise had been earlier in the month, giving the threat actor time to make preparations for the final stages of the attack.

Figure 3: Event log showing the activity of UserA from IP 45.135.187[.]108 

2. Office365: UserB 

Looking into the model breach ‘SaaS / Access / Suspicious Credential Use And Login User-Agent’, it was seen that a second account, UserB, was also observed logging in from a rare and potentially malicious location in Bangladesh [7]. Similar to UserA, this user had previously logged in exclusively from the USA, and no other accounts within the digital estate had been observed interacting with the Bangladeshi IP address. The login event appeared to bypass MFA (Multi-factor Authentication) and a suspicious user agent, BAV2ROPC, was used. Against misconfigured accounts, this Microsoft user agent is commonly used by attackers to bypass MFA on Office365. It targets Exchange’s Basic Authentication (normally used in POP3/IMAP4 conditions) and results in an OAuth flow which circumvents the additional password security brought by MFA [10].  

During the session, additional resources were accessed which appear to be associated with bill and invoice payments. In addition, on the 4th November, two new suspicious email rules named “..” were created from rare IPs (107.10.56[.]48 and 76.189.202[.]66). This type of behavior is commonly seen during SaaS compromises to delete or forward emails. Typically, an email rule created by a human user will be named to reflect the change being made, such as ‘Move emails from Legal to Urgent’. In contrast, malicious email rules are often short and undescriptive. The rule “..” is likely to blend in without arousing suspicion, while also being easy for the attacker to create and remember. 

Details of these rule changes are as follows:

·      Time: 2021-11-04 13:25:06, 2021-11-05 15:50:00 [UTC]
·      SaaS Service: Office365
·      SaaS Service Product: Exchange
·      SaaS Status Message: True
·      SaaS Source IP addresses: 107.10.56[.]48, 76.189.202[.]66
·      SaaS Account Name: O365
·      SaaS Actor: UserB@[REDACTED]
·      SaaS Event: SetInboxRule
·      SaaS Office365 Modified Property Names:
          o   AlwaysDeleteOutlookRulesBlob, Force, Identity, MoveToFolder, Name, FromAddressContainsWords, StopProcessingRules
          o   AlwaysDeleteOutlookRulesBlob, Force, Identity, Name, FromAddressContainsWords, StopProcessingRules
·      SaaS Resource Name: .. 

During cloud account compromises, attackers will often use sync operations to download emails to their local email client. During the operations, these clients typically download a large set of mail items from the cloud to a local computer. If the attacker is able to sync all mail items to their mail client, the entire mailbox can be compromised. The attacker is able to disconnect from the account and review and search the email without generating additional event logs. 

Both accounts UserA and UserB were observed using ‘MailItemsAccessed’ sync operations between the 1st and 23rd November when this attack occurred. However, based on the originating IP of the sync operations, the activity is likely to have been initiated by the legitimate, US-based users. Once the security team were able to confirm the events were expected and legitimate, they could establish that the contents of the mailbox were not a part of the data breach. 

Accomplish Mission

After gaining access to the Office365 accounts, sensitive data was downloaded by the attackers to their local system. Either on or before 14th December, the attacker had seemingly uploaded the documents onto a data leak website. In total, 130MB of data had been made available for download in two separate packages. The packages included audit and accounting financial documents, with file extensions such as DB, XLSX, and PDF.

Figure 4: The two data packages uploaded by the attacker and the extracted contents

In a sample of past SaaS activity of UserA, the subject and attachments appear related to the ‘OUTSTANDING PREPAY WIRES 2021’ excel document found from the data leak website in Figure 4, suggesting a further possibility that the account was associated with the leaked data. 

Historic SaaS activity associated with UserA: 

·      Time: 2021-11-05 21:21:18 [UTC]
·      SaaS Office365 Logon Type: Owner
·      Protocol: OFFICE365
·      SaaS Account Name: O365
·      SaaS Actor: UserA@[REDACTED].com
·      SaaS Event: Send
·      SaaS Service: Office365
·      SaaS Service Product: Exchange
·      SaaS Status Message: Succeeded
·      SaaS Office365 Attachment: WIRE 2021.xlsx (92406b); image.png (9084b); image.png (1454b); image.png (1648b); image.png (1691b); image.png (1909b); image.png (2094b)
·      SaaS Office365 Subject: Wires 11/8/21
·      SaaS Resource Location: \Drafts
·      SaaS User Agent: Client=OWA;Action=ViaProxy 

Based on the available evidence, it is highly likely that the data packages contain the data stolen during the account compromise the previous month.  

Once the credentials of an Office365 account are stolen, an attacker can not only access the user's mailbox, but also a full range of Office365 applications such as SharePoint folders, Teams Chat, or files in the user's OneDrive [11]. For example, files shared in Teams chat are stored in OneDrive for Business in a folder named Microsoft Teams Chat Files in the default Document library on SharePoint. One of the files visible on the data leak website, called ‘[REDACTED] CONTRACT.3.1.2020.pdf’, was also observed in the default document folder of a third user account (UserC) within the victim organization, suggesting the compromised accounts may have been able to access shared files stored on other accounts by moving laterally via other O365 applications such as Teams. 

One example can be seen in the below AS logs: 

·      Time: 2021-11-11 01:58:35 [UTC]
·      SaaS Resource Type: File
·      Protocol: OFFICE365
·      SaaS Account Name: 0365
·      SaaS Actor: UserC@[REDACTED]
·      SaaS Event: FilePreviewed
·      SaaS Service Product: OneDrive
·      SaaS Metric: ResourceViewed
·      SaaS Office365 Application Name: Media Analysis and Transformation Service
·      SaaS Office365 File Extension: pdf
·      SaaS Resource Location: https://[REDACTED]-my.sharepoint.com/personal/userC_[REDACTED]_com/Documents/Microsoft Teams Chat Files/[REDACTED] CONTRACT 3.1.2020.pdf
·      SaaS Resource Name: [REDACTED] CONTRACT 3.1.2020.pdf
·      SaaS Service: Office365
·      SaaS Service Product: OneDrive
·      SaaS User Agent: OneDriveMpc-Transform_Thumbnail/1.0 

In the period between the 1st and 30th November, the customer’s Darktrace DETECT/Apps trial had raised multiple high-level alerts associated with SaaS account compromise, but there was no evidence of file encryption.  

Establish Foothold 

Looking back at the start of the attack, it is unclear exactly how the attacker evaded the customer’s pre-existing security stack. At the time of the incident, the victim was using a Barracuda email gateway and Microsoft 365 Threat Management for their cloud environment. 

Darktrace detected no indication the accounts were compromised via credential bruteforcing, which would have enabled the attacker to bypass the Azure Active Directory smart lockout (if enabled). The credentials may have been harvested via a phishing campaign which successfully evaded the list of known ‘bad’ domains maintained by their email gateway.  

Upon gaining access to the account, the Microsoft Defender for Cloud Apps anomaly detection policies would have been expected to raise an alert [12]. In this instance, the unusual login from Nigeria occurred over 16 hours after the previous login from the US, potentially evading anomaly detection policies such as the ‘Impossible Travel’ rule. 

Figure 5: Event log showing the user accessing mail from USA a day before the suspicious usage from Nigeria 

Darktrace Coverage

Darktrace DETECT 

Throughout this event, high scoring model breaches associated with the attack were visible in the customer’s SaaS Console. In addition, there were two Cyber AI Analyst incidents for ‘Possible Account Hijack’ associated with the two compromised SaaS Office365 accounts, UserA and UserB. The visibility given by Darktrace DETECT also enabled the security team to confirm which files had been accessed and were likely part of the data leak.

Figure 6: Example Cyber AI Analyst incident of UserB SaaS Office365 account

Darktrace RESPOND

In this incident, the attackers successfully compromised O365 accounts in order to exfiltrate customer data. Whilst Darktrace RESPOND/Apps was being trialed and suggested several actions, it was configured in human confirmation mode. The following RESPOND/Apps actions were advised for these activities:  

·      ‘Antigena [RESPOND] Unusual Access Block’ triggered by the successful login from an unusual IP address, would have actioned the ‘Block IP’ inhibitor, preventing access to the account from the unusual IP for up to 24 hours
·      ‘Suspicious Source Activity Block’, triggered by the suspicious user agent used to bypass MFA, would have actioned the ‘Disable User’ inhibitor, disabling the user account for up to 24 hours 

During this incident, Darktrace RESPOND/Network was being used in fully autonomous mode in order to prevent the threat actor from pivoting into the network. The security team were unable to conclusively say if any attempts by the attacker to do this had been made. 

Concluding Thoughts  

Data theft extortion has become a widely used attack technique, and ransomware gangs may increasingly use this technique alone to target organizations without secure data encryption and storage policies.  

This case study describes a SaaS data theft extortion incident which bypassed MFA and existing security tools. The attacker appeared to compromise credentials without bruteforce activity, possibly with the use of social engineering through phishing. However, from the first new login, Darktrace DETECT identified the unusual credential use in spite of it being an existing account. Had Darktrace RESPOND/Apps been configured, it would have autonomously responded to halt this login and prevent the attacker from accomplishing their data theft mission.

Thanks to Oakley Cox, Brianna Leddy and Shuh Chin Goh for their contributions.

Appendices

References 

[1] https://securelist.com/new-ransomware-trends-in-2022/106457/

[2] https://www.itpro.co.uk/security/ransomware/367624/the-rise-of-double-extortion-ransomware

[3] https://www.malwarebytes.com/blog/news/2020/10/vastaamo-psychotherapy-data-breach-sees-the-most-vulnerable-victims-extorted

[4] https://www.blackfog.com/shift-from-ransomware-to-data-theft-extortion/

[5] https://www.abuseipdb.com/check/105.112.59.83

[6] https://www.abuseipdb.com/check/105.112.36.212

[7] https://www.abuseipdb.com/check/45.130.83.129

[8] https://docs.microsoft.com/en-us/microsoft-365/compliance/mailitemsaccessed-forensics-investigations?view=o365-worldwide

[9] https://www.abuseipdb.com/check/45.135.187.108

[10] https://www.virustotal.com/gui/ip-address/45.137.20.65/details

[11] https://tidorg.com/new-bec-phishing-attack-steals-office-365-credentials-and-bypasses-mfa/

[12] https://docs.microsoft.com/en-us/microsoft-365/security/office-365-security/responding-to-a-compromised-email-account?view=o365-worldwide

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.
Written by
Adrianne Marques
Senior Research Analyst

More in this series

No items found.

Blog

/

AI

/

July 13, 2026

Security After Signatures: Operating in a World of Pre‑CVE Disclosure Exploitation, Collapsed Trust Boundaries, and Autonomous Systems

Default blog imageDefault blog image

Three shifts have reshaped what it means to defend an enterprise securely.  

First, exploitation often begins before defenders have a Common Vulnerabilities and Exposures (CVE) identifier, a security advisory, or an entry in the Cybersecurity and Infrastructure Security Agency's (CISA) Known Exploited Vulnerabilities (KEV) catalog.

Secondly, the trust boundary has moved beyond the network edge into identities, tokens, APIs, and Software-as-a-Service (SaaS) workflows.  

Third, an increasing share of business activity is executed through automation, integrations, and AI agent-like systems that can act faster than teams can verify intent.  

If your security model still relies on detecting known bad artefacts, triaging isolated alerts, and waiting for confirmation before acting, you are already behind the threat.  

This is not a failure of security teams; it’s a failure of the operating model to keep pace with how the environment has changed.

A SOC built around alerts and signatures assumes that malicious activity will eventually surface as an event. In real incidents, however, the decisive evidence is rarely a single event. Instead, it is a chain of individually explainable actions that only appears malicious once you connect the dots across identity, non-human identity, cloud, email, SaaS, operational technology (OT), and network telemetry.

The defenders succeeding today observe behaviors, link them into sequences, understand what those sequences mean, and contain impact before the full story unfolds. That is the operating model the current threat environment demands.  

Exploitation before disclosure

The first shift is the straightforward: the time to exploit has dropped to nearly zero.  

In one example, Darktrace observed a sequence of subtle but strategically significant anomalies within a customer environment that later aligned with exploitation of CVE‑2025‑0994 in Trimble Cityworks by likely Chinese-nexus threat actors. Behavioral indicators were visible at least 18 days before public disclosure, with related anomalies emerging 40 to 50 days earlier during the intrusion window.  

This case illustrates a familiar pattern: clusters of weak‑signal anomalies combing to form an actionable picture of intrusion long before a CVE is published. Such activity reflects long‑horizon, option‑preserving operator models often associated with mature state‑linked activity.  

Figure 1: Darktrace’s detection of malicious exploitation of CVE 2025-0994, later tied to Chinese-nexus threat actors targeting critical national infrastructure (CNI) in the US, weeks before public disclosure.

Throughout 2025 and 2026, Darktrace has continued to observe the value of anomaly-based detections across a range of incidents.

CVE CVE Public Disclosure Date Darktrace Detection Date Days Between Detection of Exploitation and CVE Public Disclosure
CVE 2025 0994
(Trimble City Works)
2025-02-06 2025-01-19 18 Days
CVE 2025-24183
(Apache)
2025-03-10 2025-02-18 20 days
CVE 2025-10035
(Fortra GoAnywhere)
2025-09-18 2025-09-11 7 days

Identity is the real control plane

The second shift is that identity has replaced perimeter as the primary control plane. As Darktrace’s Annual Threat Report 2026 illustrated, identity remains the main challenge in defending against modern intrusions. A clear example is the Adversary-in-the-Middle (AiTM) case published by Darktrace in December 2025. A phishing email led to the compromise of an Office 365 account. Session hijacking bypassed multi-factor authentication (MFA), and the compromised account was used for follow-on phishing and persistence activities including the creation of malicious email rules.  

Every step in that sequence mattered. A successful login alone does not prove legitimacy. An inbox rule, on its own, may not appear catastrophic. Mail activity, viewed in isolation, may seem operationally normal. But the behavioral chain tells a different story: credential theft, token abuse, persistence, and onward compromise through a trusted identity.  

This is why the question is no longer “Did the user authenticate successfully”. The more important question is, “Does this identity action make sense right now, in this context, given what came before it?” The AiTM case shows how identity can be compromised. In practice, however, attacks rarely remained confined to identity alone.  

In another Darktrace case, a compromised SaaS account triggered activity across the email, SaaS, and network layers, including inbox rule changes, phishing propagation, and connections to suspicious infrastructure. Viewed in isolation, none of these events were decisive. Together, however,  they formed a behavioral sequence that revealed the intrusion, with the full attack story automatically correlated and surfaced to defenders by Darktrace’s Cyber AI Analyst.  

Figure 2: Cyber AI Analyst correlated and appended additional events to the incident, including other users who connected to the suspicious redirect link after outbound phishing emails were sent.

AI accelerates the threat  

The third shift is the one many teams still underestimate: trusted tooling, integrations, and AI agent-like systems can create actions that appear legitimate but are strategically dangerous.  

The shift becomes clearer when examining how governments are now framing AI risk. In 2026, guidance published by CISA, UK’s National Cyber Security Centre (NCSC) and Five Eyes partners warned that agentic systems expand attack surfaces, accumulate privilege, and can behave in ways that are difficult to predict or explain [1]. The advice is simple: assume unexpected behavior and design controls around it.  

The real risk is not AI usage. It is unknown autonomy: systems with credentials, data access, and action paths that can execute workflow steps without sufficient behavioral validation, traceability, or human oversight. Darktrace’s Model Context Protocol (MCP) risk analysis provides a useful framework for understanding this challenge. Over-privileged agents, content injection, and tool abuse become high-consequence risks when connected systems can dynamically retrieve data, execute actions, and communicate externally.  

Whether security teams like it or not, AI is already in the enterprise. It will help drive innovation, but it will also be abused, whether accidentally or maliciously. In each of the cases below, AI either scaled the attacker, built the tooling, or existed within the environment as something to exploit or misuse.

1. AI as an Attack Multiplier

In one campaign targeting Mexican government entities, a single operator used commercial AI platforms to generate exploits, automate reconnaissance, and process large volumes of data, compressing work that would traditionally have required an entire team into a single workflow [2].  

Darktrace is also observing this trend further down the stack. In one case, Darktrace identified AI-generated malware exploiting React2Shell, where an attacker used a Large Language Model (LLM) to produce working exploit code and deploy it at scale.  

[darktrace.com], [darktrace.com]

2. AI as an Attack Surface

Attempted AI exploitation is now appearing within customer environments. In one case involving an automation technology manufacturer, a compromised LLM proxy was seemingly used as a stepping stone to access additional AI services. When that attempt failed, the attacker pivoted to cryptomining.

What is clear is that the AI layer has already become an asset worth probing, exploiting, and pivoting through. It is also clear that defenders benefit from rapidly understanding how these activities connect. In this case, Cyber AI Analyst automatically pieced together the intrusion, while Darktrace’s Managed Threat Detection service alerted to the customer, enabling the activity to be contained before it could progress further.

Figure 3: Cyber AI Analyst's investigation into a compromised LLM proxy that was abused for cryptomining activity.

AI as a trusted but dangerous actor

This does not require a cinematic vision of “rogue AI.” The Salesloft incident provides a more grounded example, where AI and automation operate with legitimate access but served malicious intent. In that case, attackers abused compromised OAuth tokens associated with the Drift AI chat agent to export significant volumes of data from Salesforce environments.  

The activity resembled legitimate API usage and relied on trusted SaaS integrations rather than malware or other obvious signs of intrusion. That is precisely the challenge. Traditional security controls are good at detecting forced entry, but far less effective when a trusted application integration behaves in a way that is technically permitted yet operationally harmful.  

In these scenarios, the security challenge shifts from validating access to validating behavior.

This is what that looks like in practice: AI-linked identities executing legitimate actions that require behavioral validation rather than access validation.

Figure 4: Darktrace / SECURE AI highlights anomalous activity across AI identities, surfacing critical behavior that requires validation and containment.

Early observations from Darktrace / SECURE AI deployments reinforce this reality. Across Darktrace's observed fleet, AI service connections per deployment increased 13% during the first half of 2026, reaching over 16 million connections overall. The typical organisation now interacts with seven different AI providers, evidence that AI is no longer operating at the edges of the enterprise. It is increasingly woven into day-to-day business activity.

The most common risks are not compromised models or advanced AI attacks. Instead, they stem from employees and business functions exposing sensitive information through entirely legitimate-looking interactions. Darktrace has observed repeated submission of personally identifiable information (PII), tax information, identification documents, and medical data into LLM prompts, alongside widespread use of unsanctioned (shadow) AI services and growing AI activity from mobile devices.  

For defenders, the challenge is increasingly one of context: understanding when legitimate business use crosses into material risk, while preserving privacy and user trust.

Conclusion

Across all three shifts, the pattern is the same: behavior precedes understanding. Security teams are not losing because adversaries have become invisible. An increasingly outdated security model assumes that malicious activity will reveal itself cleanly and early. It no longer does.  

In 2026 and beyond, defenders win by understanding behavioral sequences, continuously validating trust, and acting before certainty becomes hindsight. That is security after signatures. That is security in the AI era.

Credit to: Daniel Levy, Threat Hunting Data Scientist

Edited by: Ryan Traill, Content Manager

References

[1] https://www.cyber.gov.au/business-government/secure-design/artificial-intelligence/careful-adoption-of-agentic-ai-services  

[2]https://www.latimes.com/business/story/2026-02-26/hacker-used-anthropics-claude-ai-to-steal-mexican-government-data

Continue reading
About the author
Nathaniel Jones
VP, Security & AI Strategy, Field CISO

Blog

/

AI

/

July 10, 2026

AIインフラがアタックサーフェスの一部に

Default blog imageDefault blog image

AIインフラとアタックサーフェスの進化

多くの組織が生成AIを実運用環境に導入するなかで、企業のクラウド環境内に新たなインフラのレイヤーが出現しています。それはAIゲートウェイです。AIゲートウェイはユーザー、アプリケーション、基盤モデルの間に位置し、多くの場合クラウドの特権アクセスを保持し、さまざまなAIサービスへのアクセスを大規模に管理しています。

AIゲートウェイとは?

AIゲートウェイはユーザー、アプリケーション、基盤モデルの間に位置し、多くの場合クラウドの特権アクセスを保持し、さまざまなAIサービスへのアクセスを大規模に管理しています。

こうした役割から、AIゲートウェイは企業のアタックサーフェスのますます重要な一部になりつつあります。AIゲートウェイが侵害されれば、攻撃者に対して計算リソースへのアクセスだけでなく、クラウドアイデンティティ、モデルサービス、機密性の高いプロンプト、そして他の接続されたシステムへのアクセスも提供してしまいます。

このブログでは、Amazon Bedrock サービスに接続されたAIゲートウェイが侵害され、その後暗号通貨マイニングインフラとの通信が観測された事例をダークトレースがどのように調査したかを解説します。問題のインスタンスは、その構成、ならびに関連するIAM(Identity and Access Management)ロールから、Amazon BedrockでホスティングされるAIサービスへのゲートウェイとして機能していることがわかりました。疑わしい侵害アクティビティが発生した後、このホストは既知の暗号通貨マイニングインフラに繰り返し通信を行い、その後シャットダウンされた様子が観測されました。Darktrace はこのアクティビティを検知し、Enhanced MonitoringおよびManaged Threat Detectionサービスを通じてエスカレーションを行いました。

この事例では最終的影響は不正な暗号通貨マイニングでしたが、このインシデントが注目に値するのはその発生場所です。侵害されたアセットは、クラウドインフラ、アイデンティティ、各種AIサービスの交差する場所に位置していました。最近の調査では、LiteLLM等のAIゲートウェイが、認証情報、モデルへのアクセス、クラウド権限を中央管理するその能力から、攻撃者にとって魅力的な標的となる可能性が明らかになっています。このアクティビティと公開されているLiteLLM脆弱性を直接結びつける証拠は見つかっていませんが、このインシデントは、AIインフラを個別のアプリケーション層として見るのではなく、重要なアタックサーフェスの一部として扱う必要性があることを表しています[1]。

暗号通貨マイニングがクラウド侵害後のアクティビティとしてよく見られる背景

暗号通貨マイニングはクラウド環境において、侵害後のアクティビティとして収益性の高いものとなり得ます。クラウド資産にアクセスできるようになった後、攻撃者はマイニングソフトウェアを展開して被害者の計算リソースを悪用し金銭的利益を得ることができます。この種のアクティビティは多くの場合機会主義的なものであり、露出したサービス、弱い認証情報、漏洩したアクセスキー、脆弱なアプリケーション、あるいはクラウドワークロードの設定ミスなどを標的として実行されます。

典型的なクラウド上での暗号通貨マイニング侵入には次のようなアクティビティが含まれます:

  • 露出したあるいは脆弱なクラウドインフラの特定
  • 露出したサービス、認証情報、またはアプリケーションの脆弱性を通じたアクセスの獲得
  • マイニングソフトウェアのダウンロードおよび実行
  • マイニングプールインフラへのアウトバウンド接続を繰り返し確立
  • アクティビティが検知され停止されるまで継続して計算リソースを消費

この事例において注目すべき要素は暗号通貨マイニングだけではありません。それが発生した場所が、AI関連アクティビティをサポートするクラウドインフラ上だったことです。この事例は、AIサービスを実現するためのアセットも、よくあるクラウド侵害リスクにさらされる可能性があることを示しています。

Amazon Bedrockに接続されたAIゲートウェイの侵害を調査

2026年6月12日、DarktraceはLiteLLM-Proxyという名前のAmazon Web Service (AWS) EC2インスタンスから暗号通貨マイニング発生中とみられるアクティビティを観測しました。このインスタンスはLiteLLMアクティビティをサポートしており、Amazon Bedrockリソースへのアクセス権を有するインスタンスプロファイルと関連付けられていました。  

AIゲートウェイは大規模言語モデルへのアクセスを中央管理するよう設計されており、多くの場合AIアプリケーションに対する認証、ルーティング、ログ、ポリシー適用を扱っています。セキュリティの視点から見ると、クラウド権限、モデルアクセス、アプリケーションワークフローを単一の制御ポイントに集約する役割も果たしています。その結果、AIゲートウェイの侵害は、侵害されたホストだけにとどまらない影響を及ぼす可能性があります。

確定的な初期アクセスベクトルは確認できませんでしたが、このアクティビティはインターネットに接続されているシステムの侵害でよく見られる次のような順序に従っていました。ブルートフォースアクセス、ペイロードの投下、そしてマイニングプールインフラに対する繰り返しのアウトバウンド接続です。

ステージ1: インターネットに露出したSSHからの初期アクセス

暗号通貨マイニングアクティビティが観測される前、LiteLLM-Proxy EC2インスタンスはSSH(ポート22)が0.0.0.0/0に対して開かれ、外部に公開されていました。

図1:EC2インスタンスがSSHポート22に対してすべてのインバウンドトラフィックを許可している設定ミスをDarktraceが警告

暗号通貨マイニングアクティビティに先立って、Darktraceはこのインスタンスに対する大量のインバウンド接続の試みが外部IPアドレス(主に145.241.123[.]102)からポート22に対して行われていることを観測しました。これはブルートフォースアクティビティを示唆するものです [2]。これらの接続の多くは短命であり、数秒しか続いておらず、スキャニングまたはログインの失敗を示していました。

図2:Darktraceがデバイスのポート22に対する不審なインバウンド接続試行を検知

入手できたテレメトリーではこれらのインバウンドSSH接続のいずれかが認証の成功につながったかどうかの確認に至らず、このアクティビティが初期アクセスベクトルであると断定することはできませんでした。しかしながら、SSHの露出、外部IPアドレスからのインバウンド接続、それに続くマイニングアクティビティは、SSHがアクセス経路の可能性が高いことを示唆しています。

ステージ2: AIゲートウェイへのXMRigマルウェアのダウンロード

最初に観測されたマイニングプールへの接続の後、このEC2インスタンスは3.42 MBのデータをポート80上のHTTP接続を介して外部エンドポイント185.62.1[.]8にダウンロードしました。このエンドポイントは暗号通貨マイニングマルウェアXMRigを含むZIPファイルをホスティングしていました[3][4]。ホストレベルのログは入手できなかったため、ダークトレースはマイニングツールがどのように実行されたか、あるいは前のSSHアクティビティがペイロード投下を直接的に可能にしたかどうかを確認できませんでした。しかしながら、ダウンロードのタイミングとその後ほどなくマイニングプールへの接続が繰り返されたことは、このインスタンスが侵害されて不正な計算アクティビティに使われたという評価を裏付けています。

ステージ3 – 侵害されたAIゲートウェイが暗号通貨マイニングインフラと通信

わずか数分後、DarktraceはLiteLLM-ProxyEC2インスタンスがHTTPs(ポート443)でホスト名pool.hasvault[.]proに対して接続していることを確認しました。最初の接続の後、同じホスト名に対して繰り返しアウトバウンド接続が観測されました。これは、侵害されたホストがマイニングインフラと通信しワークを受け取り、結果を送信するという、暗号通貨マイニングプールとの通信のパターンと一致しています。

このアクティビティがDarktraceのEnhanced Monitoringモデル“Compromise / HighPriority Crypto Currency Mining”をトリガーし、ダークトレースのSOCにより顧客に対してエスカレーションされました。また、このアクティビティはCyber AI Analystによって分析され、関連するイベントが1つの調査ナラティブにまとめられました。これにより、影響を受けたクラウドアセットからマニングプールへの繰り返しの接続を特定することができました。

図3:CyberAI Analystによる暗号通貨マイニングアクティビティの調査  

ポート443上のHTTPSの使用にも注目すべきです。なぜならば、単独で見れば、このトラフィックそのものは疑わしく見えないかもしれないからです。しかしこのケースでは、接続先、接続の量、そして類似のアクティビティが他にないことなどが、この通信を疑わしいものとして特定するのに必要な、動作のコンテキストを提供することになりました。

ステージ4: Managed Threat Detectionサービスによるリソース乱用の特定

暗号通貨マイニングアクティビティがダークトレースのManaged Threat Detectionサービスにより検知され、ダークトレースのSOCによりレビューされました。レビューの結果、このアクティビティは顧客向けにエスカレーションされました。このエスカレーションにより、顧客はAWS環境で現在発生中のリソースの乱用について、タイムリーな通知を受けることができました。

ステージ5: クラウド認証情報の不正使用とみられる疑わしいIAMアクティビティ

これとは別に、6月13日、Darktraceは別のIAMユーザーから発生した疑わしいアクティビティを検知しました。

図4: DarktraceのAdvanced Search機能が別のIAMユーザーが実行した疑わしいアクティビティをハイライト

まず、このユーザーは “GetSendQuota”イベントを試行している様子が見られました。このアクションは少なくとも過去3か月間にこのアカウントによって実行されたことのないアクションです。また、このコマンドのソースIPアドレスは14.176.1[.]47でした。地理位置情報はベトナムであり、このユーザーのアクティビティがAmazon IPアドレスから最も多く見られた場所です。さらに、このアクティビティに対してAWS CLIが使用されており、これもこのユーザーにとって通常とは異なる振る舞いでした。このことは、Darktraceの“IaaS / Unusual Activity / UnusualAWS CLI Activity”モデルによって検知されました。

図5: Darktraceによる “GetSendQuota” イベントの検知

このIAMユーザーからは、長期アクセスキーを使った疑わしいアクティビティがさらに観測されました。中でも、“InvokeModel” および “ListFoundationModels”コマンドの失敗が検知されており、モデル列挙や起動などAmazon Bedrockサービスとのやり取りを試行したことがわかります。これは前日観測されたLiteLLM侵害への関連を思わせますが、2つのイベントを確定的に結びつける証拠は不十分でした。

“CreateUser”コマンドの試行も注目に値します。なぜなら要求されたユーザー名は意味が薄いものであり、新しいアカウントを作成することにより永続性を確立する試みと見られるからです。このアクティビティはDarktraceのモデル“IaaS / Admin / New AWS UserAccount Creation”をトリガーしました。

図6:Darktraceによる“CreateUser” イベントの検知

2つのインシデント間に結びつきは確認できなかったものの、このIAMアクティビティには重要な意味があります。これは、クラウド侵害の調査においてワークロードのテレメトリーとコントロールプレーンのテレメトリーの両方を取り入れることの重要性を表しています。EC2暗号通貨マイニングアクティビティが計算リソースの乱用を示す一方、IAMアクティビティは認証情報の侵害や長期アクセスキーの不正使用、そしてクラウトサービスの不正使用の可能性を示唆しているからです。

AIインフラ保護のための重要な教訓

このインシデントの重大性は暗号通貨マイニングアクティビティそのものではなく、それが発生した場所にあります。侵害されたシステムはAmazon Bedrockサービスへのアクセス権を持つAIゲートウェイとして機能し、クラウドインフラ、アイデンティティ、そしてさまざまなAIオペレーションの交差する場所に位置していました。組織がAI機能を実運用環境に導入していくなかで、これらのプラットフォームは、露出したサービス、認証情報窃取、クラウドの設定ミスなどを通じて攻撃者がすでに狙っているアタックサーフェスの一部となりつつあるのです。

このケースでは詳細な侵入経路は特定されておらず、ワークロードの侵害と調査中に検知された疑わしいIAMアクティビティの間に決定的なつながりは確認されませんでしたが、これらのイベントは全体的な現状を裏付けています。つまり、AIインフラは個別のテクノロジースタックとして扱うのではなく、クラウド環境全体の一部として保護しなければならないとうことです。

このケースでは、最も目立った侵害の兆候は暗号通貨マイニングインフラとの通信でした。しかしここで得られたより重要な教訓は、このインシデントの全貌が理解される前にDarktraceのビヘイビア分析により明らかになった、高い権限を持つAI関連アセットを取り巻くリスクです。AIゲートウェイによりクラウド権限、モデルアクセス、アプリケーションワークフローがますます集約されるなかで、防御者は個別のアラートに集中するよりも、ワークロード、アイデンティティ、サービスの間でどのように動作がつながっているかを理解することに重点を置く必要があるでしょう。

協力:Angel Arribas Lopez (Associate Principal Cyber Analyst)、Nathaniel Jones (Field CISO/VP Threat Research)、Emma Foulger (Global Threat Ops)、Mark Turner(Security Researcher)

編集:Ryan Traill (Content Manager)

付録

Darktraceによるモデル検知結果

·       Compromise / High Priority Crypto Currency Mining

·       Compromise / Monero Mining

·       Device / Internet Facing Device with High Priority Alert

·       IaaS / Unusual Activity / Unusual AWS CLI Activity

·       IaaS / Admin / New AWS User Account Creation

MITRE ATT&CK マッピング

初期アクセス – 外部リモートサービス – T1133

初期アクセス – 有効なアカウント – T1078

実行 – コマンドおよびスクリプトインタプリタ – T1059

永続化 – アカウント作成 – T1136

探索 – クラウドサービス探索 – T1526

影響 – リソースハイジャッキング– T1496

参考資料

[1] https://docs.litellm.ai/blog/security-update-march-2026

[2] https://www.abuseipdb.com/check/145.241.123.102

[3] https://urlscan.io/search/#185.62.1.8

[4] https://www.virustotal.com/gui/file/85de36ff66fae9f4b059cbedf6d36e017ebc26c828f99f911a96e78636f21200/community

Continue reading
About the author
Angel Arribas Lopez
Associate Principal Cyber Analyst
あなたのデータ × DarktraceのAI
唯一無二のDarktrace AIで、ネットワークセキュリティを次の次元へ