The cloud has become part of the attack infrastructure itself. Threat actors can use familiar services and valid identities to control malware, move data and disrupt devices while their activity resembles ordinary enterprise traffic. For defenders, that shifts attention from suspicious files and unfamiliar servers to the API calls, permissions and administrative actions that connect cloud environments.
This shift, often described as “living off the cloud,” follows enterprise adoption of hybrid infrastructure across AWS, Azure and Google Cloud. Rather than relying only on local tools such as PowerShell or Windows Management Instrumentation, attackers can use cloud consoles, APIs, service accounts and stolen tokens. Mitiga threat hunting and response lead Arif Khan says valid credentials can let an intruder enumerate resources, extract data, raise privileges and establish persistence through routine-looking administrative calls.
That activity can be difficult to distinguish from legitimate use. Security controls built around domain reputation and static blocklists may offer little help when traffic passes through a major provider or a company’s own SaaS tenant. The practical challenge is to understand what an identity is allowed to do, and whether its actions fit its normal role.
Trusted services become command channels
Google and Mandiant disrupted a suspected Chinese espionage operation tracked as UNC2814 that used Google Sheets for command and control. Its Gridtide malware connected to an attacker-controlled spreadsheet, using a service account token to check designated cells for instructions and write task results into nearby columns. The traffic traveled through a widely used productivity service instead of an attacker-owned server.
Google’s researchers describe the technique as part of a wider move toward abusing SaaS platforms rather than building and maintaining custom infrastructure. The spreadsheet effectively served as a live control panel. For defenders, the relevant signal may be an unusual service account or unexpected access pattern, rather than a connection to a known malicious host.
Other campaigns put malicious communications behind APIs used for ordinary work. Microsoft reported that the SesameOp backdoor routed its command traffic through OpenAI’s Assistants API, making requests resemble AI development activity. Peneto Labs managing director Parthiban Jegatheesan says that kind of traffic can be hard to block without disrupting genuine workflows.
Malware including VEILDrive and malicious variants of the Havoc post-exploitation framework have also abused Microsoft Graph. According to senior offensive security engineer Kwangyun Keum, malware can authenticate to a corporate SharePoint or OneDrive tenant, retrieve command files such as cmd.txt, then place results such as results.json in a folder that looks like a user’s backup.
Identity and management access can do the damage
Cloud identity is a route to data and control even when attackers never install traditional malware on the systems they reach. In September, healthcare technology company Veradigm said attackers used a third-party vendor’s credentials to access some patient personal information, including Social Security numbers in some cases. The Gentlemen ransomware group claimed the attack and threatened to release stolen data. Veradigm said a small number of customers were affected and that clinical and medical data were not exposed.
Black Duck cybersecurity governance head Ronald Lewis characterized the incident as a case of attackers exploiting a trusted identity with access to sensitive information. Veradigm said the credentials reached a particular interface, rather than its wider network. That distinction illustrates why access reviews need to account for third-party accounts and the specific data each connected service can reach.
Administrative tools can also turn account compromise into operational disruption. In March, US medical-device maker Stryker suffered a destructive attack after attackers reportedly accessed its Microsoft Intune endpoint management platform. The Iran-linked hacktivist group Handala reportedly used that access to wipe thousands of employee laptops and mobile devices. Stryker said the incident affected operations and caused global disruption; some health systems rescheduled surgeries when the company could not deliver patient-specific implants and components.
Intune’s role in the incident underscores the power concentrated in cloud management planes. A compromised administrator account may be able to issue actions across many devices using the platform’s normal controls. Lewis said the consequences extended beyond Stryker to healthcare providers relying on its products.
Cloud storage and communications support theft
Object storage gives attackers a convenient place to stage payloads and configuration files. Rather than hosting every component on their own servers, they can retrieve a second-stage payload only when needed or change a stored file without redeploying the original malware. Peneto Labs’ Jegatheesan says this approach can reduce the malware’s footprint on a victim’s disk.
Everyday collaboration tools can also serve as collection points. Nightwing cyber incident response manager Nicholas Carroll says attackers have used Slack and Discord to move sensitive information out of compromised environments. In campaigns he described, compromised servers sent HTTPS POST requests to api.slack.com, hooks.slack.com or discord.com, delivering items such as AWS access keys, SSH keys and internal API tokens into attacker-controlled chat channels.
That choice exploits the same trust that makes these services useful at work: connections to them are common and may be permitted by network controls. Monitoring therefore needs to consider the identity, destination and content of cloud service activity, not just whether the destination is a familiar provider.
Attack chains can run across cloud infrastructure
Some operations use cloud services at multiple stages, from credential theft to data removal. CrowdStrike principal engineer Diptamay Sanyal says the China-linked group Genesis Panda has, since March 2024, queried AWS Instance Metadata Service for credentials, hosted payloads in cloud storage, used domains impersonating cloud services for command and control, and relied on cloud compute for data exfiltration. In this case, the cloud provides the operational backbone of the campaign.
Serverless infrastructure offers another way to distribute activity. During the HazyBeacon campaign targeting government entities in Southeast Asia, attackers used AWS Lambda or similar functions for reconnaissance and scanning, according to Palo Alto Networks’ Unit 42. Whisper Security co-founder Kaveh Ranjbar says an attacker can split scanning across many short-lived functions, each examining a small part of a network before stopping. Because the traffic comes from Amazon addresses that change, blocking it wholesale could also interfere with legitimate AWS use.
Cloud tunneling tools can create a persistent route through a company’s perimeter. An intruder with access to an internal server can install a Cloudflare Tunnel or ngrok agent that initiates an outbound connection, a pattern often allowed by firewalls. Ranjbar says the resulting encrypted HTTPS traffic can look like ordinary communication with a trusted provider while carrying command traffic back to the compromised system.
Attackers can also move data through cloud management APIs without transferring files over a conventional network connection. Scattered Spider and Storm-0501 have been associated with abusing AWS EBS snapshot sharing. Keum describes an attacker creating a snapshot of a victim server’s disk and sharing it with an external cloud account using the ModifySnapshotAttribute API. The attacker can then restore the snapshot and inspect it offline, including for credentials.
Trust relationships and secrets create paths between systems
Connections between cloud tenants can provide a route from a service provider into its customers. CrowdStrike says China-linked actor Murky Panda compromised upstream IT service providers and used trusted Entra ID tenant relationships to reach downstream organizations. Microsoft has also reported that ransomware group Storm-0501 targets Entra ID configurations to gain administrative privileges.
Cloud secrets stores are another direct source of access. Storm-0501 has used AWS Secrets Manager as part of broader ransomware and extortion activity. Rather than collecting credentials from endpoints, attackers can query a vault through cloud APIs, potentially avoiding endpoint-focused detection. The security value of a secrets store depends on tight permissions and monitoring of who reads its contents.
Phishing campaigns exploit trust at the point where users authenticate. Russia-linked group Cozy Bear, also tracked as APT 29, has sent victims through links that redirect to authentic Microsoft login pages. CrowdStrike’s Sanyal says victims saw Microsoft infrastructure throughout the process, undermining URL-based detection. Other phishing kits, including EvilTokens, Tycoon2FA and Kali365, have used Cloudflare’s workers.dev and similar developer platforms to host pages. ReliaQuest executive Brandon Tirado says these services can help phishing pages pass email gateways and signature checks; he also says AI can reproduce a target’s branding from a URL.
Cloud-native malware expands the threat
Some attackers have gone beyond abusing existing services and built malware for cloud environments. Check Point researchers identified VoidLink, a framework designed to compromise AWS, Azure, Google Cloud and Kubernetes infrastructure. The source describes it as including custom loaders, implants, rootkits and modular plugins intended to support persistence.
Across these examples, the common thread is the use of legitimate cloud capabilities under attacker control. A spreadsheet can carry instructions; an endpoint platform can wipe devices; a storage snapshot can expose a server’s contents. Defenders will need visibility into identities, tenant relationships, API activity and management actions, because the next suspicious event may arrive through a service the business depends on.