Application Security & Adversarial Input

← Back to Domain-Specific Families

Abstractions about injection, scripting, encryption, information flow, key custody, unsafe serialization, plaintext exposure, and attacks on learned systems.

9 abstractions in this family — domain-specific abstractions that sit near one another in structural-signature space (k-means over structural-signature embeddings). Each is shown with its short description.

• Cross-Site Scripting — Insert attacker-supplied bytes into a server's output so the victim's browser executes them under the trusted site's origin, arising when content is written into a sink without encoding for the context it lands in.
• Encryption — Transform a plaintext under a key into a ciphertext from which the message cannot be feasibly recovered without the decryption key — a computational asymmetry gated by a secret, so confidentiality depends only on who holds the key rather than on the channel.
• Information Flow (Information-Flow Security) — Track whether observations in one security domain can depend on information from another, enforcing a policy over explicit, implicit, and indirect program channels rather than only access to stored objects.
• Injection Weakness — The software-security failure in which untrusted data crosses into a control channel and a downstream interpreter executes it as command, query, or instruction — a collapse of the data/control boundary that a legitimate, often credential-free, input channel is enough to exploit.
• Insecure Deserialization — Recognize the weakness where a runtime reconstitutes an untrusted serialised byte stream into behaviour-bearing objects because the parser's own semantics execute code during the parse — safe only if the parser's power sits strictly below the runtime it feeds.
• Key Escrow — A cryptographic access-control arrangement that prepositions decryption key material with one or more custodians for release or reconstruction only under a specified recovery authorization.
• Plaintext — Identify data in the input-to-encryption or output-from-decryption role at a declared cryptographic layer, regardless of human readability, while separating that relative role from cleartext exposure and message encoding.
• Prompt Injection — Untrusted content delivered to a language model through a data channel is interpreted as instructions rather than material to process, so an embedded directive executes at the model's privilege — the failure being the absence of any in-band boundary between instructions and data, not a lapse in alignment.
• Transfer-Learning Attack — Vulnerabilities, backdoors, or poisoned representations baked into an upstream pretrained model ride intact into every downstream system built on it, because the downstream team's audit boundary encloses only its new layers while the attack surface spans the whole inherited substrate.