The core risk is this: removable media bypass every network control a government agency has paid for. A USB drive, external hard disk, or memory card carried through a front door connects an untrusted device directly to a trusted endpoint, and no firewall, intrusion-detection system, or zero-trust gateway inspects that path. The second sentence of any honest risk assessment says the same thing β the threat is physical, and physical threats need physical answers.
That's not a hypothetical. The U.S. Department of Defense and the U.K. Ministry of Defence have both restricted removable media by default for years, and the U.S. Cybersecurity and Infrastructure Security Agency runs a dedicated program β the Cyber Security Evaluation Tool, paired with its Media and Device Control guidance β because agencies keep getting hit through exactly this channel.
How a USB Port Becomes an Attack Path Into an Air-Gapped Network
A USB port is an attack path because it accepts data from a device the endpoint's operating system trusts by default. Windows, macOS, and most Linux distributions mount removable storage automatically, and until an administrator or endpoint-management tool changes that default, any device presenting a valid USB mass-storage class interface will be recognized and read.
The most damaging variant is the air-gap jump. Government networks handling classified material are often physically isolated from the public internet, which makes removable media one of the few remaining ways to move data in or out β and therefore one of the few channels an adversary can use. The Stuxnet worm, discovered in 2010, spread to Iranian nuclear facility systems at least partly through infected USB drives, and it demonstrated that an air gap is not a security boundary when someone carries a device across it.
Attackers exploit this in two directions. Inbound, a booby-trapped drive delivers malware to an isolated host. Outbound, malware already resident on the isolated host writes stolen data to any inserted drive, which a human then carries to a connected machine. The U.S. National Security Agency's published guidance on removable media calls out this bidirectional risk explicitly.
The reason this matters more than a routine phishing email is scale and persistence. A single infected drive can seed an entire enclave because the receiving machine has no reason to distrust it. That's why the U.S. Defense Information Systems Agency publishes Security Technical Implementation Guides that mandate removable-media controls on DoD systems, and why the U.K. National Cyber Security Centre's device-security guidance treats removable media as its own control family rather than a subset of endpoint protection.
The Four Distinct Risks Removable Media Pose to Government Systems
Removable media pose four separate risks, and conflating them produces weak controls. They are malware introduction, data exfiltration, hardware-level attacks that live below the operating system, and policy or supply-chain exposure when devices are procured and tracked carelessly. Each needs a different mitigation, and a program that only addresses the first will fail on the other three.
Malware introduction is the most publicized. An infected drive delivers a payload when it's mounted, sometimes before the user clicks anything, if the endpoint's autorun behavior is enabled. The U.S. Computer Emergency Readiness Team β now part of CISA β has warned for years that disabling autorun is a baseline step that many agencies still hadn't completed.
Data exfiltration is quieter and often more costly. Classified or sensitive data copied to a personal drive leaves the controlled environment entirely; once it's on a device in someone's pocket, no agency logging system follows it. The U.K. Information Commissioner's Office has fined public bodies over lost unencrypted storage media, and the pattern repeats because the loss often isn't discovered until the device turns up in the wrong hands.
Hardware-level attacks are the least visible. A malicious USB device can present itself as a keyboard and issue commands in milliseconds β the technique known as a BadUSB attack, described publicly by security researchers Karsten Nohl and Jakob Lell in 2014 β or it can contain hidden storage that the operating system never enumerates. Because the attack lives in firmware, reformatting the drive doesn't remove it.
Supply-chain and policy exposure rounds out the set. An agency that buys cheap unbranded drives, or that can't say which drives are authorized, has no inventory to audit and no way to detect a substituted device. The U.S. National Institute of Standards and Technology's Special Publication 800-53 lists media protection as a control family precisely because procurement and tracking are security functions, not administrative ones.
Why Air-Gapped Government Networks Are Not Immune to Media-Borne Attacks
Air-gapped networks are vulnerable because isolation stops network attacks, not human ones. Every transfer into or out of an isolated enclave requires a person and a device, and that person is the attack surface. The 2010 Stuxnet operation is the canonical case, but the pattern has recurred in defense and critical-infrastructure environments repeatedly since.
The U.S. Government Accountability Office has reported on federal agencies' incomplete implementation of media-protection controls across multiple audits, noting that agencies often lack complete inventories of removable media and can't demonstrate that devices are scanned before use. That gap is the practical vulnerability: the policy exists, the enforcement doesn't.
Insider risk compounds it. A cleared employee with legitimate access can carry data out on a drive that looks identical to an authorized one, and unless the agency runs device-level allowlisting, the system will accept it. The U.K. National Cyber Security Centre's insider-risk guidance notes that removable media are one of the most common vectors in insider data-loss cases precisely because the action is fast and hard to reverse.
There's also the cross-domain problem. Some agencies legitimately need to move data between classification levels, and removable media are one approved method when handled through a controlled transfer process. The risk isn't the transfer β it's that the same ports and the same drives are available for unapproved transfers, and distinguishing the two requires technical enforcement, not just procedure.
A Worked Table: Which Control Blocks Which Attack
The table below maps each of the four risks to the control that actually stops it. These controls appear in NIST SP 800-53's MP (Media Protection) control family and in DISA's STIGs, though agencies implement them unevenly.
| Risk | Primary control | What it enforces | Common failure |
|---|---|---|---|
| Malware introduction | Device allowlisting via endpoint policy | Only approved device IDs mount at all | Allowlist not updated; legacy drives still work |
| Data exfiltration | Read-only or write-blocked ports for unapproved devices | Data can't be copied out | Users plug into unmanaged machines |
| Firmware-level attack | Hardware-enforced port disablement | Ports physically off when not needed | Software toggle only; BIOS override possible |
| Supply-chain exposure | Centralized procurement and asset tagging | Every authorized drive is inventoried | Personal drives mixed with issued ones |
Read the table as a checklist, not a menu. The allowlisting row is the highest-leverage control because it stops three of the four risks at the point of insertion, but it only works if the allowlist is managed centrally and the endpoint can't be reconfigured by the user. DISA's removable-media STIG requires exactly that kind of enforcement, and it also requires logging every mount event so that an unauthorized attempt is visible after the fact.
The write-blocking row matters most in environments where read-only transfers are legitimate. If a machine only ever needs to read briefing material from a drive, configuring its ports to deny writes removes the exfiltration path without breaking the workflow. That's a design decision, not a policy one, and it has to be made per machine class rather than agency-wide.
The firmware row is where software-only programs break down. A BadUSB device reflashes itself to look like a keyboard, and no operating-system-level control will stop it once the device is recognized as an input device rather than storage. Hardware-enforced port disablement β physical port blockers, BIOS-level lockouts, or portless thin-client configurations β is the only reliable answer, which is why high-security enclaves often remove ports entirely.
Step-by-Step: Hardening Endpoints Against Media-Borne Intrusion
Follow these steps in order. They assume a Windows endpoint managed through Group Policy or an equivalent endpoint-management console; macOS and Linux have functional equivalents under different names, and the DISA STIG for removable media names the specific registry and policy settings on Windows.
1: Disable autorun and autoplay for all drive types. In Group Policy, set "Turn off Autoplay" under Computer Configuration β Administrative Templates β Windows Components β AutoPlay Policies to "All drives." The default is "Not configured," which leaves autorun active for many device classes. 2: Apply a device-installation allowlist. Under Device Installation Restrictions, set "Prevent installation of removable devices" to enabled, then add approved device IDs as exceptions. Only drives on the exception list will mount. 3: Set removable storage to read-only by default. Under Removable Storage Access, deny write access to removable disks, then grant write permissions only to specific approved device classes. 4: Enable mount auditing. Turn on Object Access auditing for removable storage so every insertion and file copy is logged to the Security event log and forwarded to the agency's SIEM. 5: Physically block unused ports. Install port blockers or apply epoxy to ports that have no approved use case; this is the only control that survives a firmware-level attack. 6: Inventory every authorized drive. Tag each approved device, record its serial number, and reconcile the inventory quarterly against the allowlist. 7: Test the failure path. Insert an unauthorized drive and confirm it doesn't mount, that the attempt is logged, and that the endpoint doesn't fall back to a permissive default.
The failure-handling close matters as much as the steps. If an unauthorized drive does mount, treat the endpoint as compromised: isolate it from the network, image it for forensics, and don't simply reformat the drive and return it to service. If the allowlist blocks a legitimate device, add it through the central console rather than granting a local exception, because local exceptions are how allowlists quietly become denylists.
The Program-Level Costs Agencies Often Underestimate
Removable-media control carries real program costs, and agencies that budget only for software licenses miss most of them. The visible line item is endpoint management tooling, which for a mid-size agency runs into six figures annually depending on seat count and vendor. The larger cost is labor: inventory reconciliation, exception handling, and forensic response to incidents all consume staff time that rarely appears in the initial estimate.
There's a usability tax too. When ports are locked down, legitimate workflows slow down. Field staff who used to copy briefing files to a personal drive now have to request an approved device, wait for issuance, and log the transfer. The U.K. National Cyber Security Centre's guidance on usable security makes the point that controls users route around are controls that don't work, so the program has to provide a fast approved path or it will be defeated by workarounds.
Hardware costs are modest but real. Port blockers cost a few dollars each; physical portless thin clients cost more per seat than standard desktops. The trade-off is worth it in high-classification enclaves and rarely worth it in low-risk administrative environments, which is why agencies tier their controls by machine class rather than applying one standard everywhere.
Recovery costs dominate the tail. A single confirmed media-borne compromise can trigger full enclave reimaging, forensic investigation, and in classified environments a damage assessment that runs for months. The U.S. Department of Defense's incident-handling requirements under its cloud and on-premises security frameworks make that assessment mandatory, and the cost of one incident typically exceeds years of preventive control spending.
Comparing Removable-Media Controls Across Three Common Agency Environments
The right control set depends on the environment, and the three configurations below cover most government endpoints. Each row names the control that carries the most weight and the trade-off that comes with it.
| Environment | Dominant control | Strength | Trade-off |
|---|---|---|---|
| Classified air-gapped enclave | Ports physically removed or blocked; no removable media permitted | Eliminates the channel entirely | Breaks all media-based workflows; requires alternative transfer process |
| Controlled unclassified network | Device allowlist plus read-only default plus audit logging | Blocks most attacks while preserving approved transfers | Requires central management and quarterly inventory discipline |
| Administrative or low-risk endpoints | Disable autorun, scan on mount, user awareness training | Low cost, fast to deploy | Doesn't stop determined insiders or firmware attacks |
The classified enclave wins on security and loses on flexibility. Removing ports is the only control that survives a determined adversary with physical access, but it forces every legitimate transfer through a separate controlled process β typically a dedicated transfer system or a data diode β which costs money and adds latency to routine work.
The controlled unclassified network is where most agencies actually operate, and the allowlist-plus-read-only-plus-audit combination is the balanced answer. It preserves approved workflows, stops the most common attacks, and produces the logs needed for investigation. Its weakness is management discipline: an allowlist that isn't reconciled quarterly drifts into a denylist.
The administrative tier is where many agencies start, and it's the weakest. Disabling autorun and scanning on mount are worthwhile baseline steps, but they don't stop a BadUSB device or an insider with an authorized-looking drive. Agencies should treat this tier as a floor, not a ceiling, and move higher-risk endpoints up to the controlled tier even when the classification doesn't strictly require it.
Why Malware Scanning Alone Doesn't Solve the Problem
Malware scanning on mount is necessary but insufficient, and treating it as the primary control is the most common program error. Scanners only catch known signatures and heuristics, so a freshly compiled or firmware-resident payload passes through. The scanning step also happens after the device is already recognized by the operating system, which means the attack surface has already expanded.
The timing problem is structural. A BadUSB device issues its keystrokes before any scanner runs, because the scanner waits for a mounted filesystem and the device never presents one. A scanner can't inspect what it can't see, and firmware-level attacks are invisible to file-level scanning by design.
Signature lag compounds it. New malware families often have a window of hours to days before scanners update, and a targeted attack against a specific agency can be compiled and delivered inside that window. The U.S. National Institute of Standards and Technology's malware-incident guidance notes that signature-based detection is a layer, not a boundary.
The practical conclusion is that scanning belongs in the stack but not at the top. Allowlisting, read-only defaults, and physical port control do the heavy lifting; scanning catches what slips through those and provides a detection signal for the SIEM. An agency that scans but doesn't allowlist has the controls in the wrong order.
Enforcement Gaps That Let Media-Borne Attacks Succeed
Enforcement gaps, not missing policies, are where most media-borne attacks succeed. Almost every government agency has a removable-media policy on paper; the failures happen when the policy isn't technically enforced and users find the path of least resistance. The U.S. Government Accountability Office's repeated findings on incomplete media-protection implementation point at exactly this gap.
The first gap is unmanaged endpoints. A contractor laptop or a legacy machine outside the endpoint-management console won't receive the allowlist policy, so its ports stay open. Agencies often discover these machines only during incident response, which is the worst possible time.
The second gap is local administrator rights. A user with admin access can re-enable a disabled port, override a policy, or install a driver that circumvents controls. DISA's STIGs require removing local admin rights from standard users for this reason, and agencies that haven't done so have a policy that depends on user cooperation.
The third gap is exception sprawl. Every approved exception is a hole, and agencies rarely close exceptions when the need expires. Over a few years, a tight allowlist becomes a patchwork of one-off approvals that no one reviews, and the effective control is weaker than the documentation suggests.
FAQ
Can antivirus software stop a USB drive from infecting a government computer?
Antivirus alone cannot stop it, because the infection can occur before the scanner runs. A BadUSB device issues commands as a keyboard and never presents a filesystem for scanning, so no antivirus product will see it. Disabling autorun plus device allowlisting closes that gap; scanning then catches what gets through.
What is the single most effective control against removable-media attacks?
Physically blocking or removing unused USB ports is the most effective single control, because it survives firmware-level attacks that defeat software controls. Agencies that can't remove ports should pair device allowlisting with read-only defaults, which stops three of the four risk categories at insertion. The trade-off is workflow disruption.
Are government air-gapped networks safe from USB attacks?
No. Air gaps stop network attacks but leave removable media as one of the few remaining channels, and Stuxnet demonstrated the consequence in 2010. Every transfer into an isolated enclave involves a person and a device, and the person is the attack surface. Air-gapped enclaves typically remove ports entirely rather than rely on policy.
How much does a removable-media control program cost an agency?
Tooling for a mid-size agency runs into six figures annually, plus staff time for inventory reconciliation and incident response. Port blockers cost a few dollars each, while portless thin clients cost more per seat than standard desktops. One confirmed compromise typically exceeds years of preventive spending because of reimaging and damage assessment.
What should happen when an unauthorized USB drive is inserted?
Treat the endpoint as compromised: isolate it from the network, image it for forensics, and do not simply reformat the drive and return it to service. Reformatting doesn't remove firmware-level malware. The insertion attempt should be logged and investigated even if the device failed to mount.
Do these controls apply to contractors and personal devices?
Yes, and that's where most programs break. Contractor laptops outside the endpoint-management console don't receive allowlist policy, so their ports stay open. Requiring contractor devices to enroll in management before network access, and banning personal drives outright, closes the gap that agencies most often discover during incident response.
What is BadUSB and why does it defeat normal security tools?
BadUSB is a class of attack where a device's firmware is reprogrammed to impersonate a keyboard or network adapter instead of storage. Security researchers Karsten Nohl and Jakob Lell demonstrated it publicly in 2014. Because the operating system sees an input device rather than a drive, storage-specific controls never engage.
This article is general educational information about security risk, not a diagnosis, treatment plan, legal opinion, or investment recommendation. It cannot substitute for an assessment by a qualified security professional familiar with your agency's systems and applicable requirements. If you're responding to a suspected compromise, contact your organization's incident-response team and, where national-security systems are involved, the relevant authorities immediately.