{
  "id": "malware",
  "title": "Malware",
  "lang": "en",
  "source_lang": "en",
  "summary": "Malware is software designed to compromise systems, steal information, disrupt operations, or perform other unauthorized actions.",
  "category": "computer-science",
  "subcategory": "networks-security",
  "type": "concept",
  "aliases": [
    "Malicious software",
    "malicious code"
  ],
  "generated_at": "2026-11-02",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/malware/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/malware/",
      "json": "https://aiwiki.page/zh/malware.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Malware is software or code intentionally designed to perform unauthorized actions that compromise the confidentiality, integrity, or availability of computing systems and their data. It includes programs that replicate between machines, steal credentials, encrypt files for extortion, secretly monitor users, or provide remote control to an intruder. Malware can operate on personal devices, enterprise servers, industrial equipment, and network infrastructure. Its defining characteristic is malicious purpose, not a particular programming language, file format, or method of execution.\n\nMalware is a central subject of cybersecurity because it converts weaknesses in software, system configuration, identity management, and human decision-making into operational capabilities. A single malicious program may combine several functions, while a prolonged intrusion may involve many programs alongside legitimate administrative tools. Understanding malware therefore requires attention both to code and to the wider environment in which that code is delivered, executed, controlled, detected, and removed."
    },
    {
      "heading": "Definition, boundaries, and classification",
      "text": "Malware is defined primarily by intended behavior and authorization. A program that secretly copies confidential documents to an outside party is malicious even if it contains no self-replication mechanism and exploits no software defect. Conversely, a program that deletes files is not necessarily malware: deletion may be the authorized purpose of a maintenance utility. Context distinguishes legitimate administration, testing, and monitoring from actions performed deceptively or without permission.\n\nThis distinction creates a boundary around dual-use software. Remote administration programs, password-recovery utilities, debuggers, and network scanners have legitimate applications but can also support intrusions. Security products sometimes classify such software separately as potentially unwanted applications or riskware. These labels describe concerns about consent, installation practices, or capabilities rather than proving that every use is malicious. Intrusive advertising software, for example, may range from openly disclosed functionality to covert tracking or interference with browser settings.\n\nMalware classification uses several independent dimensions. One concerns propagation: whether a program attaches to another file, independently spreads across networks, or depends on installation by a user. Another concerns its effect, such as surveillance, extortion, sabotage, or resource theft. Other dimensions include the targeted operating system, the execution layer, and the mechanism used to maintain access. These categories overlap. A single specimen may be a self-propagating program, a credential stealer, and a vehicle for installing additional payloads.\n\nA sample is a particular collected artifact, such as an executable file, script, or document. A variant is a related version, while a family groups programs judged to share meaningful code ancestry or design. A campaign is an operational activity that may employ several unrelated families. Security vendors often assign different names to the same family, and superficially similar names do not necessarily indicate a common origin. File hashes identify exact byte sequences, not the full set of related programs.\n\nAn exploit and malware are also distinct concepts. An exploit takes advantage of a vulnerability to produce behavior that the system was not intended to permit. It may serve as an entry mechanism for malware, but many malicious programs arrive through ordinary installation or stolen accounts. Conversely, an exploit can appear in authorized security research without being used maliciously. A buffer overflow is one vulnerability class that can enable exploitation, but malware does not require memory corruption.\n\nThe broader information security objective is protection of systems and information, whereas malware analysis concerns one important class of threats. Fraud conducted entirely through misleading messages, account misuse, or authorized interfaces may cause serious harm without deploying malware at all."
    },
    {
      "heading": "Major forms and their relationships",
      "text": "A computer virus replicates by inserting or attaching its code to a host object, such as an executable file, document, or boot sector. When the host is processed or executed, the virus gains an opportunity to run and potentially infect other hosts. Replication is its defining property; visible damage is not required. Some viruses preserve the apparent operation of infected programs, while others corrupt them accidentally or deliberately.\n\nA computer worm is a self-replicating program that spreads between systems without needing to attach itself to a host program in the way a conventional file virus does. Worms may exploit exposed services, copy themselves through shared storage, or use messaging systems. Their operational impact can arise from their payload, from the network traffic generated during propagation, or from resource exhaustion on infected machines. The distinction between viruses and worms becomes less clear in programs that combine both mechanisms.\n\nA Trojan horse presents itself as useful or harmless while carrying out concealed malicious actions. Unlike a virus or worm, it need not replicate. Fake installers, modified utilities, and applications that secretly capture credentials fit this category. A dropper places another malicious component onto a system, while a downloader retrieves additional components from elsewhere. These functional labels can apply to the same program at different stages of an infection.\n\nRansomware attempts to coerce payment by denying access to systems or data, commonly through encryption. Some operations also steal information and threaten disclosure. A wiper, by contrast, is designed to destroy data or render systems unusable rather than provide a practical recovery path after payment. The outward appearance can be deceptive: destructive malware may display a ransom demand even when reliable restoration was never part of its design.\n\nSpyware covertly collects information about a user or organization. Its capabilities can include recording keystrokes, taking screenshots, accessing messages, or activating sensors. Credential stealers specialize in passwords, authentication tokens, browser session information, and related secrets. Banking malware may manipulate a financial session or intercept information rather than merely copy stored passwords. These distinctions describe the collection objective, not necessarily separate code bases.\n\nA rootkit conceals malicious activity or preserves privileged access by altering the behavior of a system. A bootkit targets the startup process. A backdoor provides an access path that bypasses normal controls, while a remotely controlled bot can become part of a botnet. Such terms describe different aspects of operation: a single infection can contain a bootkit for persistence, a rootkit for concealment, and a bot component for remote tasking. Treating malware categories as mutually exclusive therefore obscures how actual infections are assembled."
    },
    {
      "heading": "Historical development",
      "text": "Malware emerged from experiments with self-reproducing computation and from the increasing interconnection of computers. The Creeper program, developed by Bob Thomas at BBN in 1971, is commonly discussed as an early experimental precursor to network worms. It was not equivalent in purpose or operation to modern criminal malware. Early examples demonstrate that replication and movement between machines existed as technical ideas before a large commercial security industry developed around them.\n\nIn 1983, Fred Cohen conducted experiments with self-replicating programs, and his subsequent work helped establish the computer virus as a subject of security research. During the 1980s, removable disks were a major distribution channel. Elk Cloner, associated with Apple II computers in 1982, and Brain, an IBM PC boot-sector virus that appeared in 1986, became prominent early examples. Infection often depended on exchanging disks and starting a computer from infected media.\n\nThe Morris worm of November 1988 marked a major episode in the security history of the Internet. Released by Robert Tappan Morris, it affected Unix systems through multiple mechanisms, including weaknesses in network services and password guessing. Repeated infections imposed substantial load on affected machines. The incident contributed to the creation of the CERT Coordination Center at Carnegie Mellon University and demonstrated how rapidly connected systems could experience a shared operational emergency.\n\nThe expansion of email, office automation, and the World Wide Web changed distribution during the 1990s and early 2000s. Macro viruses used the scripting capabilities of office documents. Melissa in 1999 combined document-based infection with email distribution, while ILOVEYOU in 2000 used a script attachment and deceptive presentation. Code Red in 2001 and SQL Slammer in 2003 exploited network services, showing that widespread disruption could occur without each victim opening a malicious attachment.\n\nBy the mid-2000s, financial incentives increasingly shaped malware development. Credential theft, unsolicited messaging, advertising fraud, and botnet rental supported specialized services and criminal markets. The Zeus banking Trojan became a prominent example of malware organized around financial theft. Stuxnet, publicly identified in 2010, illustrated a different objective: manipulation of industrial control processes. Its targeting of Siemens control-system technology made the relationship between malicious software and physical equipment especially visible.\n\nLater incidents exposed the consequences of interconnected infrastructure and software distribution. Mirai in 2016 recruited connected devices into a botnet associated with major denial-of-service attacks. WannaCry in May 2017 combined ransomware with worm-like propagation. NotPetya in June 2017 caused destructive international disruption after an initial distribution route involving compromised accounting-software updates. The SolarWinds compromise, disclosed in 2020, demonstrated how malicious code introduced into a trusted software build and update process could facilitate access to numerous organizations."
    },
    {
      "heading": "Delivery and initial execution",
      "text": "Malware delivery is the process by which malicious code reaches an environment; initial execution is the point at which the environment begins interpreting or running it. These are separate events. A malicious attachment may remain inert in a mailbox, and a downloaded executable may never run. Conversely, exploitation of a remotely reachable service may cause execution without creating a conventional downloaded file or requiring interaction by the device’s user.\n\nPhishing is a common delivery context because a message can persuade a recipient to open an attachment, follow a link, or install purportedly necessary software. Attachments may contain executable code directly or use document features that invoke scripts and external content. The precise chain depends on application behavior and security settings. A deceptive login page, however, is not itself evidence of malware: credential theft can occur entirely through a website that records submitted information.\n\nWeb-based delivery includes compromised sites, malicious advertisements, and counterfeit downloads. In a drive-by compromise, merely visiting or interacting with web content can expose a vulnerable browser or component to exploitation. Modern web browsers limit this through process isolation, permission systems, and application sandboxes, but vulnerabilities can sometimes allow those boundaries to be crossed. Other attacks avoid exploitation and instead persuade a person to install an extension, update, or application that is falsely presented as legitimate.\n\nA software supply-chain attack compromises software through a development, dependency, packaging, or distribution relationship. Malicious code may enter a vendor’s build process, an update package, or a public package repository. The victim receives it through a channel normally associated with legitimate software. Open development does not inherently cause such attacks, but open-source software ecosystems make dependency identity, maintainer access, and package provenance especially important subjects of scrutiny. Closed-source distribution channels can be compromised as well.\n\nRemovable storage and local access remain relevant, particularly for systems that have limited network connectivity. Infection can depend on executable files, document processing, startup behavior, or a device that presents itself as a different class of peripheral. Network shares also provide a bridge between local access and enterprise-wide propagation. Malware does not invariably spread automatically from these locations; sometimes an additional user action or an already compromised account is necessary.\n\nInitial access can also precede malware deployment. An intruder may first obtain access through stolen credentials, an exposed remote service, or a configuration error, then upload a malicious program. Consequently, the first malware detected on a system may not reveal the original point of entry. Incident reconstruction distinguishes the delivery route, the first execution mechanism, and later movement between systems rather than assuming that they are a single event."
    },
    {
      "heading": "Execution, privilege, persistence, and concealment",
      "text": "After execution, malware operates within the permissions and technical constraints of its environment. A program running as an ordinary user can still access that user’s documents, browser data, and network resources. Administrative privileges expand the available actions but are not a prerequisite for meaningful harm. Privilege escalation describes the acquisition of permissions beyond those initially available, through exploitation, configuration weaknesses, stolen credentials, or misuse of trusted services.\n\nPersistence allows malicious activity to resume after a process exits, a user signs out, or a system restarts. Mechanisms vary by platform. On Microsoft Windows, malware may abuse services, scheduled tasks, or startup-related configuration. On Linux, examples include altered service definitions, login scripts, and scheduled jobs. These facilities are legitimate parts of the operating system; their malicious use depends on what is configured and who authorized it.\n\nSome infections target earlier stages of startup. A modified boot loader or compromised firmware component may execute before the main operating system and can be harder to inspect using tools within that system. Unified Extensible Firmware Interface security mechanisms, including Secure Boot, are intended to constrain unauthorized startup code. Their effectiveness depends on implementation, key management, configuration, and the integrity of trusted components. Firmware persistence is technically significant but should not be assumed to be present in ordinary infections without evidence.\n\nConcealment can occur at several levels. A malicious program may use an inconspicuous filename, place data in a legitimate-looking directory, or execute inside another process. More intrusive rootkits can manipulate system interfaces so that files, processes, or network connections are omitted from ordinary views. Process injection allows code to run in the address space of another process, but the presence of a familiar process name alone does not establish that its behavior is trustworthy.\n\nThe term fileless malware describes techniques that rely heavily on memory, scripts, system configuration, or legitimate interpreters rather than a conventional malicious executable stored on disk. It does not mean that an attack leaves no evidence or never touches persistent storage. “Living off the land” similarly refers to abuse of existing system utilities and administrative capabilities. Such activity can reduce the usefulness of detection based only on suspicious executable files.\n\nMalware may also alter its representation to frustrate inspection. Packing compresses or transforms a program and restores executable content at runtime; encryption conceals embedded code or configuration until needed. Polymorphic malware changes its visible representation between instances, while metamorphic techniques rewrite code structure while preserving behavior. These methods complicate matching, but they do not eliminate observable consequences such as unusual execution relationships, memory changes, or communications."
    },
    {
      "heading": "Command, coordination, and operational objectives",
      "text": "Many malicious programs communicate with an external controller after installation. Command and control, often abbreviated C2, provides instructions, configuration updates, additional payloads, or a route for returning results. Not all malware requires it: a wiper may act autonomously, and a self-contained ransomware program may encrypt files without maintaining an interactive connection. The presence or absence of remote tasking affects both the attacker’s flexibility and the defender’s opportunities for disruption.\n\nCentralized C2 uses one or more servers to coordinate infected systems. Peer-to-peer designs distribute communication among participants and can make removal of a single server less decisive. Some malware uses domain-generation algorithms to derive changing rendezvous names, while other programs rely on compromised websites or legitimate online services. These mechanisms complicate discovery of the controlling infrastructure but also create patterns that analysts may study.\n\nMalicious communications frequently use ordinary network protocols. Requests carried over web traffic can resemble legitimate application activity, and encryption can hide message contents from passive inspection. Transport Layer Security protects the communication channel regardless of whether the application’s purpose is legitimate. Nevertheless, metadata such as destination, timing, traffic volume, and relationships between processes and connections can remain informative. Abuse of the Domain Name System can provide another communication or data-transfer channel.\n\nInformation theft involves selecting, collecting, and exporting data. Malware may target files, mailboxes, browser stores, financial records, or secrets used by applications. Theft of session tokens can sometimes allow account access without replaying a password, although the result depends on token validity and other controls. Authentication mechanisms and device access therefore intersect: compromising an endpoint can expose information after it has legitimately been decrypted or made available to an authenticated application.\n\nBotnets aggregate the resources of many infected devices. Their tasks can include unsolicited messaging, automated fraud, proxying traffic, or denial-of-service attacks. In a distributed denial-of-service attack, many sources contribute traffic or requests that exhaust a target’s capacity. Resource theft also includes unauthorized cryptocurrency mining, where the operator consumes another party’s computing time and electricity. The direct symptoms may be elevated load rather than stolen documents or visibly corrupted files.\n\nOperational models frequently divide labor. One group may distribute malware, another purchase access, and another conduct extortion or data theft. Malware-as-a-service and ransomware-as-a-service arrangements separate software provision from deployment, sometimes with revenue sharing. These relationships mean that identifying a malware family does not by itself identify a single operator. Code can be sold, leaked, copied, or reused across unrelated activities, while one operator can employ multiple families."
    },
    {
      "heading": "Platforms and deployment environments",
      "text": "Malware is not confined to desktop computers. Its practical design reflects processor architecture, operating-system interfaces, application permissions, and deployment practices. Personal computers provide rich user data and interactive sessions, while servers provide application credentials, business records, and access to other systems. A program compiled for one platform does not automatically run on another, although scripts, document formats, and cross-platform runtimes can bridge some differences.\n\nOn smartphones, application isolation and controlled installation channels affect malware behavior. A malicious smartphone application may misuse permissions, exploit a platform vulnerability, or deceive a user into granting powerful accessibility or management capabilities. Platform security models restrict access to other applications’ data, but these boundaries can be weakened by implementation flaws or excessive privileges. Mobile surveillance can also involve compromised accounts and backup services, which should be distinguished from malware resident on the device.\n\nServers and cloud computing environments introduce different persistence and access patterns. A web shell is malicious server-side code that exposes remote command functionality through a web application. Compromised cloud credentials may permit deployment of malicious workloads without an endpoint exploit. Containerized environments share important host resources, including the operating-system kernel, so an application container is not equivalent to a separate physical machine or a fully isolated security boundary.\n\nMalware targeting a virtual machine can act entirely inside the guest system, attack shared infrastructure, or attempt to exploit a virtualization boundary. These are different threat models. Similarly, compromise of a cloud account, a virtual server, and the provider’s underlying infrastructure are distinct events. The technical and organizational scope of an incident depends on which identities and isolation layers have actually been crossed.\n\nConnected appliances, routers, cameras, and other embedded devices often have limited interfaces and long deployment lives. Their compromise may support traffic interception, botnet participation, or access to adjacent networks. Some malware on such devices runs only in volatile memory and disappears after a restart, but a still-exposed device may be reinfected. Other infections modify persistent storage or configuration. The absence of a desktop-style file browser or security interface can make inspection difficult.\n\nIndustrial environments add consequences beyond information loss. Software may communicate with programmable controllers, engineering workstations, and monitoring systems that influence physical processes. Stuxnet is a prominent example of malicious logic designed around a particular industrial context. Such targeting requires knowledge of equipment and process behavior; ordinary enterprise malware can also disrupt industrial operations indirectly by disabling scheduling, communications, or operator workstations. Direct manipulation of physical processes and disruption of supporting business systems are therefore separate forms of impact."
    },
    {
      "heading": "Analysis and detection",
      "text": "Malware analysis seeks to determine what a suspicious artifact does, how it operates, and what evidence it leaves. Static analysis examines code and data without executing the program in its intended manner. Analysts may inspect file structure, imported functions, embedded strings, digital signatures, and instructions recovered through disassembly. Decompilation can provide a higher-level approximation, but it does not reconstruct the original source code exactly.\n\nDynamic analysis observes execution in a controlled environment. A sandbox can record process creation, file changes, configuration updates, memory behavior, and network activity. Its results describe what happened under the tested conditions, not every possible behavior. Malware may require a specific document, user action, date, geographic setting, or response from a controller. It may also detect signs of analysis and suppress its payload.\n\nMemory analysis can reveal unpacked code, decrypted configuration, injected components, and active connections that are difficult to recover from stored files alone. Reverse engineering often combines these observations with knowledge of operating-system internals and application programming interfaces. Network analysis provides another perspective by identifying communication patterns and relationships among affected hosts. No single artifact necessarily explains the full intrusion.\n\nDetection methods range from exact matching to behavioral inference. Cryptographic hashes efficiently identify previously catalogued files, but changing even one byte normally changes the hash. Signatures can match broader byte patterns or structural features. Heuristic methods assess characteristics associated with malicious programs, while behavioral systems evaluate actions such as suspicious persistence changes or unusual access to credential stores. These methods differ in their resistance to evasion and their tendency to generate false alarms.\n\nMachine learning can classify files, sequences, or network events using patterns learned from labeled or unlabeled examples. Its performance depends on data quality, representative evaluation, and changes in software and attacker behavior. A model may learn incidental features of a collection rather than durable characteristics of maliciousness. Distribution shift, class imbalance, and adversarial adaptation complicate claims based only on laboratory accuracy.\n\nOperational evaluation must consider the rarity of malicious events. If malware prevalence is \\(p\\), detector sensitivity is \\(s\\), and the false-positive rate is \\(f\\), then the positive predictive value under those assumptions is\n\n\\[P(\\text{malware}\\mid\\text{alert})=\\frac{sp}{sp+f(1-p)}.\\]\n\nWith \\(p=0.001\\), \\(s=0.99\\), and \\(f=0.01\\), only about 9 percent of alerts correspond to malware in this simplified model. Real systems use additional context and correlated evidence, but the example explains why a seemingly small false-positive rate can still produce substantial investigative workload."
    },
    {
      "heading": "Containment, recovery, and evidential limits",
      "text": "Malware incident response distinguishes containment, eradication, and recovery. Containment limits further access or damage, for example by restricting communication or separating affected systems from shared resources. Eradication addresses malicious components and the conditions that allowed their operation. Recovery restores services and data while establishing whether the environment is sufficiently trustworthy to resume operation. These activities may overlap, particularly when essential services must continue during an investigation.\n\nRemoving a detected file is not equivalent to eliminating a compromise. An intruder may have created accounts, changed configuration, stolen credentials, or installed additional components. Malware may also have altered records or application data without leaving a readily identifiable executable behind. For that reason, recovery decisions depend on the scope of access and the confidence available in the system’s remaining state, not simply on whether a scanner reports no known malware.\n\nBackups can restore availability after encryption or destruction, but they do not reverse disclosure of stolen information. Their usefulness depends on integrity, accessibility, retention, and whether they contain an already compromised state. Rebuilding systems from known software and restoring data are distinct operations. A restored service may still be exposed to the original intrusion route if the associated vulnerability, credential compromise, or trust relationship remains unchanged.\n\nDigital forensics reconstructs activity from evidence such as event logs, memory captures, files, authentication records, and network telemetry. Investigators distinguish direct observations from inference and preserve provenance when evidence may be used in legal proceedings. Missing records do not establish that an event did not occur: logging may never have been enabled, retention may have expired, or an intruder may have altered evidence. Conversely, a shared filename or network address alone is rarely sufficient to establish a complete chain of events.\n\nAttribution operates at several levels. Analysts may link samples through shared code, associate infrastructure with a campaign, or assess the organization responsible for an operation. Each step requires additional evidence. Code reuse, compromised servers, deliberately misleading artifacts, and shared criminal services complicate attribution. Technical similarity does not automatically establish common authorship, and identifying a malware developer is not the same as identifying everyone who deployed the program.\n\nThere are also fundamental limits to universal detection. In general, determining arbitrary program behavior encounters undecidability results related to the halting problem, although practical security systems work with restricted representations and observable behavior rather than solving an unrestricted mathematical problem. Formal verification can establish specified properties under explicit assumptions, but it does not automatically prove that an entire deployed environment is free from malicious intent or misuse. Malware defense consequently combines imperfect technical detectors, constrained execution environments, evidence collection, and controls over software and identity."
    }
  ],
  "links": [
    "cybersecurity",
    "operating-system",
    "information-security",
    "ransomware",
    "internet",
    "email",
    "world-wide-web",
    "phishing",
    "web-browser",
    "open-source-software",
    "microsoft-windows",
    "linux",
    "encryption",
    "transport-layer-security",
    "domain-name-system",
    "authentication",
    "denial-of-service-attack",
    "personal-computer",
    "smartphone",
    "cloud-computing",
    "virtual-machine",
    "router",
    "source-code",
    "sandbox-computer-security",
    "application-programming-interface",
    "machine-learning",
    "distribution-shift",
    "halting-problem",
    "formal-verification"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}