/threat-modeling

Bug Bounty Patterns 2024-2026 — threat-modeling addendum

Overview

STRIDE-per-element extension of the existing stride_threat_library.md, digesting the 38 post-2023 bug-bounty patterns from the other skills into system-modelling language. This file is an addendum — not a replacement — and mirrors the existing library's ## <Element> / ### <STRIDE letter> shape so consumers can merge it in directly.

Source patterns (by ID) are numbered identically to the per-skill reference files; see ../../<skill>/references/bounty_patterns_2024_2026.md for operational detail.

External Entity

Spoofing

• Malicious mobile app registers same URL scheme / Android intent filter as legitimate app, spoofing the OAuth callback receiver (iOS P35, Android P34).
• Attacker-controlled Terraform Cloud workspace spoofs the organization principal accepted by an AWS OIDC trust (IaC P26).
• Attacker AI tool description spoofs a trusted MCP tool to pull instructions into a coding agent (LLM P8 / SAST P8).

Repudiation

• Refresh tokens continue minting access tokens after password reset, breaking the "user revoked access" audit narrative (API P4).
• Key-cycling makes "per-user anomalous usage" metrics unreliable (API P5).

Process

Spoofing

• JWT request_uri JAR injection swaps an attacker's identity claims into the IdP's flow (API P2).
• Unauthenticated kube-apiserver admits any caller as system:anonymous (IaC P28).

Tampering

• Prototype pollution overrides Object.prototype and tampers with library-internal allow-lists (DAST PA / SAST PA).
• Mass assignment rewrites role/permissions via ordinary PATCH endpoints (API P6).
• Helm chart drift tampers dev securityContext into prod (IaC P27).
• Malicious npm preinstall tampers with source tree during install (SCA P30).
• Git arbitrary file write via .gitmodules tampers with arbitrary paths (SCA P32).

Repudiation

• Adaptive LLM defense evasion defeats the very "judge" log meant to provide forensics (LLM P10).

Information Disclosure

• Indirect prompt injection via retrieved docs exfiltrates session data via tool calls (LLM P7).
• Multimodal injection extracts text the user never saw (LLM P9).
• ORM filter DSLs leak adjacent tenant data through boolean-timing oracles (API PB / SAST PB).
• SSRF via DNS rebinding leaks IMDS credentials (Cloud P15).
• HTTP request smuggling reveals internal vhosts and cross-user bodies (DAST P21 / Network P21).
• Cache deception serves prior user's authenticated payload to anonymous callers (DAST PC).
• Unprotected broadcast receivers leak auth tokens to any installed app (Android P37).

Denial of Service

• Cycling API keys bypasses per-key throttles, enabling distributed DoS (API P5).
• Shai-Hulud 2.0 payload wipes $HOME, a targeted DoS on developer machines (SCA P30).

Elevation of Privilege

• OAuth non-happy-path flow → account takeover (API P1).
• JWT forgery via leaked .env secret → admin ATO (API P3).
• ConfusedFunction → project-wide GCP privesc (Cloud P11).
• AWSMarketplaceFullAccess → full AWS admin via EC2 role (Cloud P14).
• Compute-SA-reassignment + network-tag chain → lateral to admin VM (Cloud P13).
• runC escape → container breakout + host compromise (Container P17).
• NVIDIA Container Toolkit LD_PRELOAD escape (Container P18).
• K8s RoleBinding creation → cluster-admin (Container P20).
• SA token theft → lateral pivot + API-server admin ops (Container P19 / Network P19).
• GitHub Action tj-actions tag retargeting → stolen CI secrets → downstream privesc (SCA P31).

Data Store

Tampering

• RAG vector / doc store poisoning embeds instructions into the knowledge base (LLM P7).
• Keychain / Keystore items with loose ACL allow on-device modification (iOS P38).

Information Disclosure

• Exposed .env / .git / backup files reveal JWT secrets, AWS keys, etc. (API P3).
• iOS backup + Keychain accessibility = off-device credential leak (iOS P38).
• Oracle EBS SSRF chain reads IMDS / internal services, disclosing secrets (Cloud P16).

Denial of Service

• Transitive dependency CVE in data-layer code (e.g., protobuf parser) reachable from network input → DoS (SCA P33).

Data Flow

Tampering

• HTTP request smuggling mutates requests between front-end and back-end (DAST P21 / Network P21).
• HTTP/2 CONNECT punches an unsanctioned data flow into internal ranges (DAST P22 / Network P22).
• WAF parser-discrepancy lets malicious bodies reach the app unmutated (DAST P23).
• Base64-encoded SSRF in headers evades WAF inspection (DAST P25).

Information Disclosure

• Blind-SSRF redirect-loop oracle turns invisible flows into measurable ones (Cloud PD).
• Cache deception splits a single authenticated flow into many anonymous ones (DAST PC).

Cross-cutting (maps to any element)

Supply-chain meta-threats

• Shai-Hulud 2.0 worm (SCA P30), tj-actions compromise (SCA P31), Git CVE-2025-48384 (SCA P32), and MCP-server tampering (LLM P8) jointly represent the "compromised upstream artefact" class — should appear on every DFD as an External-Entity-Tampering arrow into any Process that executes third-party code.

Governance / shift-left

• IaC shift-left correlation (IaC P29) is a control-maturity finding; include in every threat model as a pre-condition for mitigation credibility.

Using this addendum

1. Merge entries into your existing DFD's element table; the STRIDE letter and element type are the natural join key.
2. Cross-reference each row with ../../<skill>/references/bounty_patterns_2024_2026.md for detection/mitigation detail.
3. For DREAD scoring, assume severity ratings from the source pattern index (Critical ≥ 8, High ≥ 6) unless contextual factors warrant re-scoring.
4. Include "Supply-chain meta-threats" as a standalone section of every threat model that involves third-party libraries, AI agents, or GitOps pipelines.