OWASP Top 10 (2025) - CEH Cheatsheet Addendum

Scope: Supplement to 99-cheatsheet.md §4 (OWASP Top 10 - 2021). Read this after the 2021 table; this file covers only what changed for 2025, with the AI/LLM angle CEH v13 emphasises.

1. Status - when 2025 was released, what changed

Status: General Availability (GA / officially released), 2025. Verified at https://owasp.org/Top10/2025/ on 2026-09-11. The Introduction opens “Welcome to the 8th installment of the OWASP Top Ten!”; the footer reads © Copyright 2021-2025. The /Top10/ root now redirects to /Top10/2025/. Not RC or draft - it supersedes 2021 as the live industry baseline.

OWASP’s motivation: 8 of 10 categories are data-driven from ~175k CVE/CWE mappings across ~2.8M tested apps; two come from the community survey. The CWE pool grew from ~400 (2021) to 589; each 2025 category caps at 40 CWEs.

Headline changes: two new categories (A03 expanded, A10 new), one consolidation (SSRF folded into A01), two renames (A07, A09). Six other names unchanged; only ordering shifts.

2. New / renamed 2025 categories (table)

2025 ID	Name	Brief description	2021 ancestor
A01	Broken Access Control	AuthZ: IDOR, privilege escalation, tenant isolation. Now absorbs SSRF.	A01:2021 + A10:2021 SSRF
A02	Security Misconfiguration	Jumps #5 → #2. Default creds, verbose errors, misconfigured cloud (S3/Blob/IMDS), missing headers, debug mode in prod.	A05:2021 (renumbered)
A03	Software Supply Chain Failures	NEW / expanded. Compromise in build → distribute → update: malicious deps (Codecov, xz-utils), unsigned updates (SolarWinds), CI/CD tampering.	Expansion of A06:2021 Vulnerable & Outdated Components
A04	Cryptographic Failures	Drops #2 → #4. Weak hashing, plaintext storage, missing TLS.	A02:2021 (renumbered)
A05	Injection	Drops #3 → #5. SQLi, NoSQLi, LDAP, OS-cmd, SSTI, XSS.	A03:2021 (renumbered)
A06	Insecure Design	Drops #4 → #6. Missing threat models, business-logic flaws, no rate limits on credential flows.	A04:2021 (renumbered)
A07	Authentication Failures	Renamed from “Identification and Authentication Failures”. 36 CWEs. Push for phishing-resistant MFA.	A07:2021 (rename)
A08	Software or Data Integrity Failures	Trust-boundary failures below supply chain: unsigned deserialisation, no CI artifact integrity.	A08:2021 (same)
A09	Security Logging & Alerting Failures	Renamed from “Security Logging and Monitoring Failures”. Alerting is now non-optional.	A09:2021 (rename)
A10	Mishandling of Exceptional Conditions	NEW. 24 CWEs. Improper error handling, failing open, race conditions, null-deref on adversarial input.	None - net new

Numbering: A01, A06, A08 keep their number; A04, A05 slide down by two; A02, A03, A07, A09, A10 are new or moved. SSRF (old A10:2021) no longer has its own slot - it lives inside A01.

3. Renamed / merged categories

Change	2021	2025	Why it matters for CEH
Renamed	A07 Identification and Authentication Failures	A07 Authentication Failures	“Identification” was redundant. Memorise the new name only.
Renamed	A09 Security Logging and Monitoring Failures	A09 Security Logging & Alerting Failures	“Monitoring” was too passive; OWASP stresses alerting. CEH language follows.
Merged	A10 SSRF	Folded into A01 Broken Access Control	SSRF re-classified as server-side authZ failure (app trusted an attacker URL). File SSRF under A01.
Expanded	A06 Vulnerable and Outdated Components	A03 Software Supply Chain Failures	Scope grew from “known-CVE components” to “any compromise in build/distribute/update”.
New	-	A10 Mishandling of Exceptional Conditions	Class OWASP considers chronically under-tested: failing-open error paths, payment-race conditions.

4. AI / LLM-specific risks (the exam-fresh material)

Critical clarification: the OWASP Top 10 2025 contains no LLM/AI category. OWASP treats app security and AI security as separate projects. The companion list is the OWASP Top 10 for LLM Applications (2025) - a distinct project at https://owasp.org/www-project-top-10-for-large-language-model-applications/. CEH “OWASP AI risks” or “LLM security” questions refer to that list, not this one.

OWASP Top 10 for LLM Applications 2025: LLM01 Prompt Injection - direct + indirect (most-tested); LLM02 Sensitive Information Disclosure - training-data regurgitation, RAG doc leaks; LLM03 Supply Chain - poisoned pre-trained models, tampered fine-tuning data, malicious LoRA/tokenizers; LLM04 Data and Model Poisoning - backdoors/biases via adversarial training data; LLM05 Improper Output Handling - LLM text flows into eval(), os.system(), SQL, shell - classic injection reborn; LLM06 Excessive Agency - agentic LLMs granted tools (browser/code-exec/payments) without least-privilege; 2025 splits into excessive functionality, permissions, autonomy; LLM07 System Prompt Leakage - exposed prompt reveals hidden instructions and API shapes; LLM08 Vector and Embedding Weaknesses - RAG stores enable indirect prompt injection, embedding inversion, cross-tenant leakage; LLM09 Misinformation - hallucination, fabricated citations, unsafe completions; LLM10 Unbounded Consumption - DoS via massive prompts or expensive tool calls; model-theft via systematic API probing.

CEH v13 framing: the exam splits AI threats into AI-powered attacks (adversaries using LLMs offensively - phishing generation, malware coding, deepfake social engineering) and AI-target attacks (the LLM01-LLM10 list). Canonical attack names: indirect prompt injection, prompt extraction, many-shot jailbreak, model stealing, embedding inversion, tool poisoning.

5. CEH exam-relevant deltas (what to memorise that wasn’t in 2021)

A02 is now Top-2. Configuration taxonomy: default creds, verbose errors, directory listing, sample apps in prod, missing headers, weak cloud-IAM. Classic stem: “default-configured web server, missing patch.”
A03 is the highest-impact new entry. Three incidents: SolarWinds (2020), Codecov (2021, tampered Bash uploader exfiltrated CI secrets), xz-utils (2024, malicious maintainer). Defences: SBOM, SLSA levels, signed releases, reproducible builds, Sigstore.
SSRF has no standalone ID - A01 sub-case now. Memorise Capital One (2019): misconfigured WAF → SSRF → IMDSv1 → STS creds → 700-bucket over-privilege. Controls: IMDSv2, egress allow-list, segmentation.
A10 - Mishandling of Exceptional Conditions. Canonical CEH scenario: payment endpoint returns HTTP 200 + {success: false} on DB timeout → merchant resubmits → double-charge. Or: auth gate fails open on a malformed JWT. Know CWE-754 and CWE-755.
A07 / A09 wording shifts. A07 prefers phishing-resistant MFA (FIDO2, WebAuthn, passkeys) over TOTP/SMS. A09 makes alerting non-optional - great logs with no alert are useless.
AI/LLM is its own addendum. OWASP Top 10 2025 ≠ OWASP LLM Top 10. If the stem says “LLM” or “GenAI,” switch to LLM01-LLM10 above. Most-tested: LLM01 prompt injection and LLM06 excessive agency.

6. Sources

OWASP Foundation. OWASP Top 10:2025 - Home / List. https://owasp.org/Top10/2025/ (verified GA, 2026-09-11).
OWASP Foundation. OWASP Top 10:2025 - Introduction (“What’s changed”). https://owasp.org/Top10/2025/0x00_2025-Introduction/ - primary source for renames, ordering, SSRF consolidation, dataset size.
OWASP Foundation. A03:2025 Software Supply Chain Failures. https://owasp.org/Top10/2025/A03_2025-Software_Supply_Chain_Failures/ - 5 CWEs, CWE-477/1104/1329/1395.
OWASP Foundation. A10:2025 Mishandling of Exceptional Conditions. https://owasp.org/Top10/2025/A10_2025-Mishandling_of_Exceptional_Conditions/ - 24 CWEs.
OWASP GenAI Security Project. OWASP Top 10 for LLM Applications (2025). https://owasp.org/www-project-top-10-for-large-language-model-applications/ - LLM01-LLM10.
van der Stock, Glas, Smithline, Janca, Gigler - Lead Authors, OWASP Top 10:2025 (listed on Introduction page).
Codecov. Post-mortem of the April 2021 Bash Uploader incident. https://about.codecov.io/apr-2021-post-mortem/ - canonical supply-chain case.
Capital One breach technical write-up (SSRF → IMDSv1). https://techearl.com/capital-one-breach-ssrf ; cross-referenced with research/05-domain-3c-app-crypto-cloud.md §3.C.5.