Compromise by Extension

Expanded Definition

Compromise by extension describes a failure mode in which a single compromised NHI or AI agent exposes every system, dataset, and workflow it can reach. In NHI security, the identity is not the whole attack surface. Its permissions, delegated tokens, API scopes, and runtime tool access define the true blast radius. This becomes especially important when an AI agent is allowed to call services on behalf of users or systems, because the model may act as the interface while the attached credentials perform the actual damage.

Definitions vary across vendors on whether this is treated as a distinct control problem or as a consequence of overprivilege, but the operational meaning is consistent: compromise propagates through authorization edges. The concept aligns closely with least privilege, secret hygiene, and Zero Trust Architecture as described in NIST Cybersecurity Framework 2.0. The most common misapplication is assuming the model layer is the only asset at risk, which occurs when teams ignore the permissions carried by the connected service account, vault token, or orchestration role.

Examples and Use Cases

Implementing controls against compromise by extension rigorously often introduces operational friction, requiring organisations to weigh agent autonomy and workflow speed against tighter scope limits, token rotation, and approval gates.

• A support AI agent has read access to customer records and write access to ticketing workflows, so a prompt injection can turn a benign query path into data exfiltration and unauthorized case updates.
• A CI/CD service account can deploy code, read secrets, and manage cloud roles, so one stolen token can extend into release tampering, credential theft, and infrastructure changes.
• A data assistant is granted broad warehouse access for convenience, then inherits lateral reach into adjacent schemas and analytics exports that were never intended for the original use case.
• The risk pattern is visible in the breach history collected by The 52 NHI breaches Report, where compromised identities repeatedly became force multipliers.
• As Anthropic’s report on AI-orchestrated cyber espionage shows, an agent with tool access can be repurposed into an execution layer rather than just a text interface.

Why It Matters in NHI Security

Compromise by extension is one of the clearest reasons NHI governance cannot stop at inventorying identities. When 97% of NHIs carry excessive privileges, as noted in the Ultimate Guide to NHIs, the likely outcome of compromise is not a single account event but a chain reaction across data stores, pipelines, and control planes. That is why the same guide’s finding that 80% of identity breaches involved compromised non-human identities is so operationally important to practitioners.

Security teams need to understand this term because incident scope, containment strategy, and recovery time all depend on whether the compromised identity had narrow or expansive reach. The right response is usually to reduce standing access, isolate tool permissions, and validate every downstream dependency before restoring service. For governance, this also means pairing why NHI security matters now with explicit authorization boundaries rather than assuming one control fixes the whole chain. Organisations typically encounter compromise by extension only after an alert, outage, or data exposure reveals how far the identity could really reach, at which point the term becomes operationally unavoidable to address.

Related resources from NHI Mgmt Group

• How do I respond to a confirmed NHI credential compromise?
• How do attackers turn a supply-chain incident into wider NHI compromise?
• How should organizations manage browser extension risks?
• When should enterprises review their extension policies?