Product Security Bad Practices, Version 2, is voluntary joint guidance published in January 2025 by the Cybersecurity and Infrastructure Security Agency (CISA) and the Federal Bureau of Investigation (FBI) — "the authoring organizations," as the document calls them — under CISA's Secure by Design initiative. It catalogs 13 specific practices the two agencies call "exceptionally risky," each with a named weakness class (CWE) and a recommended action, aimed at "software manufacturers who develop software products and services, including on-premises software, cloud services, and software as a service (SaaS)," and explicitly extended to "software products that run on operational technology (OT) products or embedded systems." It is distributed as TLP:CLEAR, meaning unrestricted public disclosure, and is non-binding: "this document imposes no requirement" to follow it.
Key takeaways
- The document sorts its 13 practices into three categories that it defines itself, and only practices 1 to 4 carry a CWE ID in the text.
- It is the only item with a calendar deadline, and it touches the whole codebase.
- At release, no KEV.
- Synthetic example.
What are the 13 bad practices, and what does each mean for a device maker?
The document sorts its 13 practices into three categories that it defines itself, and only practices 1 to 4 carry a CWE ID in the text. In the tables below, the second column is CISA's bad practice. The third is the engineering change that typically closes the gap on an embedded Linux or RTOS product, plus the record worth keeping as evidence.
Product properties (8 practices)
| # | Bad practice (CWE) | Device-maker fix and evidence |
|---|---|---|
| 1 | New critical-infrastructure/NCF product lines in a memory-unsafe language (e.g., C/C++) without justification; existing memory-unsafe products without a published memory safety roadmap (CWE-119) | Write new network-facing daemons and parsers in a memory-safe language such as Rust; publish the memory safety plan with priority components |
| 2 | User-provided input placed directly into raw SQL query strings (CWE-89) | Bound parameters in the web UI's SQLite or config-database calls (OWASP cheat sheet); keep the static-analysis rule |
| 3 | User-provided input placed directly into raw OS command strings (CWE-78) | Replace system() calls that build ping, NTP or hostname commands with argument arrays or library calls (OWASP) |
| 4 | Shipping with default, universally-shared passwords (CWE-1392, CWE-1393) | Per-device password set at the factory, or password creation forced at first boot; keep the provisioning test |
| 5 | Shipping with a component already listed on CISA's Known Exploited Vulnerabilities (KEV) Catalog | Match the release SBOM against the KEV feed as a release gate; keep the match result per release |
| 6 | Shipping open-source components with a disclosed CVSS 9.0+ vulnerability | Machine-readable SBOM per build, CVE and end-of-life monitoring per component |
| 7 | Using known-insecure or deprecated cryptography (TLS 1.0/1.1, MD5, SHA-1, DES) or no encryption for sensitive data in transit/at rest | Build the TLS library without versions below 1.2; replace MD5/SHA-1 image digests with SHA-256 |
| 8 | Hardcoded credentials or secrets in source code | Move signing keys and cloud tokens to a secret manager; secret scanning on every merge |
Security features (2 practices)
| # | Bad practice | Device-maker fix and evidence |
|---|---|---|
| 9 | No native or federated multi-factor authentication (including phishing-resistant MFA) in the baseline product; MFA not enabled by default for administrators | MFA on the cloud portal and remote maintenance accounts; a published threat model where on-device MFA is unsafe |
| 10 | Insufficient logging to reconstruct common intrusion types: configuration changes, identity/sign-in and network flows, and data access or creation | Machine-readable export (syslog or structured logs) of config changes and logins |
Organizational processes and policies (3 practices)
| # | Bad practice | Device-maker fix and evidence |
|---|---|---|
| 11 | Not issuing timely, complete CVEs (with the CWE field populated) for critical (CVSS ≥ 9.0) or high-impact (CVSS ≥ 7.0) vulnerabilities | Become a CNA or work with one; fill the CWE field on every record |
| 12 | No published vulnerability disclosure policy (VDP) covering the product | Publish a VDP and security.txt with CISA's four elements |
| 13 | For on-premises products, not clearly communicating the support period | Print the support end date on the box, datasheet and product page |
The document is explicit that this list "is focused and does not include every possible inadvisable cybersecurity practice" — leaving a practice off the list is not CISA or the FBI endorsing it.
For default passwords, two of the five options CISA lists fit hardware well: "Requiring physical access for initial setup and the specification of instance-unique credentials" for devices without a screen, and "Conducting campaigns or offering updates that transition existing deployments from default passwords" for the installed base. The weakness class is CWE-1392, "The product uses default credentials (such as passwords or cryptographic keys) for potentially critical functionality." For cryptography, NIST's SP 800-52 Rev. 2 says government servers "shall not use TLS 1.0, SSL 3.0, or SSL 2.0", a useful benchmark when a customer asks why TLS 1.1 was removed. Fix guides: memory corruption, command injection, hard-coded credentials and weak cryptography.
Why is the memory safety roadmap the hardest item?
It is the only item with a calendar deadline, and it touches the whole codebase. Bad practice #1 has two parts: starting a new product line for critical-infrastructure/NCF use in a memory-unsafe language like C or C++ "where readily available alternative memory-safe languages could be used" is itself flagged as dangerous; separately, for existing products already written in memory-unsafe languages, the absence of a published memory safety roadmap is flagged the same way.
The recommended action sets a concrete date: software manufacturers should publish a memory safety roadmap by the end of 2025, prioritizing "network-facing code or code that handles sensitive functions like cryptographic operations" first. The obligation has a carve-out — it "does not apply to products that have an announced end-of-support date that is prior to Jan. 1, 2030."
The guidance's own note accepts that a full rewrite is unrealistic in the short term. It suggests writing new components in memory-safe languages and, in parallel, implementing "hardware or compiler controls to mitigate memory safety vulnerabilities," then rewriting high-risk components such as cryptographic code over time. The document names no specific controls. For embedded C, the usual candidates are stack protectors, _FORTIFY_SOURCE, position-independent executables with ASLR, and hardware memory protection, which line up with SSDF task PW.6.1 in NIST SP 800-218: "Use compiler, interpreter, and build tools that offer features to improve executable security."
What do the KEV and critical-vulnerability rules require after release?
They apply on the day you ship and every day after:
At release, no KEVA product should not ship with "a component that contains an exploitable vulnerability present on CISA's Known Exploited Vulnerabilities (KEV) Catalog." A footnote covers KEVs published within 30 days before a scheduled release: patch within 30 days of the upstream fix. After release, 30 daysWhen a new KEV affects the product, issue a patch "at no cost to its users in a timely manner (no longer than 30 days from the date of which a patch for the component containing the KEV is made available)" and warn users of the risk of not installing it. Not exploitable? Say so in publicIf the KEV "is only exploitable via a function that is never called," publish written documentation explaining why. A footnote adds that this should ideally be machine-processable, through Vulnerability Exploitability eXchange (VEX). Critical open-source CVEsPractice 6 defines critical as "a Common Vulnerability Scoring System (CVSS) score of 9.0 or greater" and asks for an SBOM "in an industry-standard, machine-readable format" provided to customers, plus routine monitoring for CVEs and end-of-life.
The KEV JSON feed listed 1,779 entries in catalog version 2026.11.10 and keeps growing, so the 30-day clock needs monitoring, not a one-off scan. See KEV and EPSS prioritization, firmware SBOM and CSAF and VEX for device makers.
Worked example: how would a camera maker triage a release?
Synthetic example. The product, versions and findings are invented to show the method; they describe no real company or device.
An IP camera maker checks firmware 4.2 against the 13 items:
SBOM and component match (practices 5, 6)The build SBOM lists BusyBox, OpenSSL, lighttpd and a vendor RTSP library. One OpenSSL CVE scores CVSS 9.1 and none of the components appears on the KEV list. OpenSSL is upgraded before release. Passwords (4, 8)The factory image ships a shared adminpassword. The team switches to a per-unit password printed on the label and forces a change at first login. A secret scan moves a cloud API token from a build script to the CI secret store.Crypto (7)TLS 1.0 is still enabled for an old video-management client. It is disabled, and the release notes name the minimum client version. Injection (3)The diagnostics page builds a pingcommand from user input. It is rewritten to pass an argument array with an IP-address allowlist.Memory safety (1)The RTSP parser is the highest-exposure C component. The memory safety roadmap names it first, with compiler hardening now and a Rust rewrite as the second step. Process (11, 12, 13)The company publishes a VDP and security.txt, states the support end date on the datasheet, and arranges CVE assignment through a CNA.
Result: six items needed work before release; MFA on the cloud portal and log export got owners and dates for the next one.
Which bad practices are already law under PSTI or the CRA?
Several. The CISA/FBI text is voluntary, but the same outcomes bind products sold in the UK or the EU:
| Bad practice | UK PSTI (in force since 29 April 2024) | EU Cyber Resilience Act |
|---|---|---|
| 4 Default passwords | Schedule 1 para 1: passwords "must be— (a) unique per product; or (b) defined by the user of the product" | Annex I Part I(2)(b): "secure by default configuration" |
| 5, 6 Known vulnerabilities | — | Annex I Part I(2)(a): "without known exploitable vulnerabilities"; Part II(1): an SBOM |
| 7 Cryptography | — | Annex I Part I(2)(e): encryption "by state of the art mechanisms" |
| 12 VDP | Schedule 1 para 2: published "information on how to report security issues" | Annex I Part II(5): "a policy on coordinated vulnerability disclosure" |
| 13 Support period | Schedule 1 para 3: published "information on minimum security update periods" | Art. 13(8): "at least five years"; Art. 13(19): end date with "at least the month and the year" at purchase |
The CRA text is on EUR-Lex. Practice 13 asks only for a clear statement "at the time of sale," with no minimum length. The CRA sets the five-year floor. See PSTI statement of compliance and the Cyber Resilience Act guide.
What changed from the October 2024 draft?
Version 1.0 was an initial public draft released in October 2024 alongside a Request for Information (Federal Register docket CISA-2024-0028). The Federal Register notice was published on 16 October 2024 and asked for comments by 2 December 2024. The draft did not mention OT or embedded systems; Version 2 does. CISA says it incorporated feedback from 78 public comments into Version 2.0, published January 2025. The net changes:
- Three brand-new bad practices: insecure/deprecated cryptographic algorithms (#7), hardcoded credentials or secrets (#8), and unclear on-premises support periods (#13) — none of these appeared in the October 2024 draft.
- Clarified KEV patch timelines (see the KEV section above).
- OT-specific MFA language: recognizing that MFA can itself introduce safety risk on some OT/medical devices (the document's own example is "medical devices in emergency rooms where delay in physician access could lead to patient harm"), manufacturers of those products should instead publish a threat model showing how they mitigate single-factor credential abuse, while still supporting MFA on lower-risk interfaces like vendor/maintenance accounts and remotely accessible engineering workstations and HMIs.
- Phishing-resistant MFA called out explicitly as the expected bar, not just any second factor. Passkeys "meet this definition even if they are the sole form of authentication"; the FIDO Alliance calls them "phishing resistant."
How does this relate to other Secure by Design artifacts?
Bad Practices isn't a standalone list — nearly every item footnotes a cross-reference to CISA's Secure by Design pledge (pledge goals are cited in practices #1, #2, #3, #4, #5, #9, #10, #11 and #12) and to specific practices (PW) and response (RV) tasks in NIST's Secure Software Development Framework (SSDF), such as PW.5.1 for injection prevention and RV.1.3 for CVE/VDP transparency. Reading the three together — principles (Secure by Design), lifecycle practices (SSDF), and concrete anti-patterns (Bad Practices) — gives a fuller picture than any one document alone.
What are the common pitfalls?
- Treating "no CWE" as fine. The CVE Program's CNA Operational Rules v4.0 say a record "SHOULD use" CWE. CISA goes further and calls a missing CWE field dangerous. See CNA rules.
- Mixing up the two exemption dates. End of support before 1 January 2030 exempts the memory safety roadmap; before 1 January 2028 exempts admin MFA by default. Nothing else is exempted.
- Reading log retention as universal. "At least 6 months" applies to cloud service providers and SaaS. For on-device logs, the ask is machine-readable export of the listed event types.
- Fixing new units only. Default passwords and KEV components in the installed base need an update campaign, which the text asks for explicitly.
- Saying "not exploitable" privately. The guidance asks for public written documentation, ideally VEX.
Where does a firmware component scan help, and where does it not?
A component scan answers practices 5 and 6 and little else. SoftWSP extracts an SBOM from a firmware image as CycloneDX 1.6 JSON, matches the components against NVD, CISA KEV, FIRST EPSS and CISA CSAF ICS advisories, and drafts a cited report that an engineer reviews before it goes out. That gives evidence for "no KEV at release" and "no CVSS 9.0+ open-source component," and an outdated OpenSSL or mbed TLS version shows up as a component finding relevant to practice 7. Watchlists alert on a CVE id or a vendor/product pair.
It does not search firmware for passwords, keys or TLS settings, review source code for injection, judge a memory safety roadmap, write VEX, or analyse whether vulnerable code is reachable. See Device & Firmware Security and the OWASP firmware testing methodology for the wider test scope.
Where should you start?
Triage against the 13 items as a checklist, not a scoring rubric — the document assigns no severity scale beyond "dangerous"; close the cheap gaps first (a published VDP and support-period statement cost little; a memory safety roadmap takes real engineering time). Start the memory safety roadmap now if you ship C/C++ firmware— the end-of-2025 publication deadline has passed as of this writing, so a late memory safety plan is itself a visible gap. Check your CVE records for the CWE field, and publish the VDP with the four elements CISA lists; the VDP template covers them. Don't treat "voluntary" as "ignorable"— default passwords, a VDP and a support-period statement are already law under PSTI and the EU Cyber Resilience Act.
This guide summarizes CISA and the FBI's own published Product Security Bad Practices, Version 2 (January 2025); it is non-binding guidance, not a regulation, and this summary is not a substitute for the full document.
Put this guide to work
Frequently asked questions
What is the CISA/FBI Product Security Bad Practices catalog?
Product Security Bad Practices is joint, voluntary guidance from the Cybersecurity and Infrastructure Security Agency (CISA) and the Federal Bureau of Investigation (FBI) — referred to in the document itself as "the authoring organizations" — published under CISA's Secure by Design initiative. Per its own Overview, it is "intended for software manufacturers who develop software products and services, including on-premises software, cloud services, and software as a service (SaaS)" and explicitly "also applies to software products that run on operational technology (OT) products or embedded systems." It lists 13 practices the authoring organizations consider "exceptionally risky," each paired with a recommended action, particularly for manufacturers whose software is "used in service of critical infrastructure or national critical functions (NCFs)." The document is distributed as TLP:CLEAR (unrestricted public disclosure) and states plainly: "The guidance contained in this document is non-binding, and while the authoring organizations encourage avoiding these bad practices, this document imposes no requirement to do so."
What changed between the October 2024 draft and the January 2025 Version 2?
Per the document's own Change Record, Version 1.0 was published in October 2024 as an initial draft, and Version 2.0 followed in January 2025 after CISA "received 78 public comments" during the Request for Information period referenced in the Federal Register docket (CISA-2024-0028). Version 2.0's changes include: three entirely new bad practices (use of known insecure or outdated cryptographic functions, hardcoded credentials or secrets in source code, and failure to clearly communicate a product's support period); additional context in the memory-safety section; more worked examples for preventing SQL injection and OS command injection; clarified patch timelines for Known Exploited Vulnerabilities (KEVs); new operational-technology-specific language for multi-factor authentication (MFA); and an added expectation that manufacturers support phishing-resistant MFA specifically, not just any second factor.
What are the 13 bad practices, and how are they grouped?
The catalog groups its 13 items into three categories it defines itself: "Product properties, which describe the observable, security-related qualities of a software product" (8 items — new products in memory-unsafe languages without a published memory safety roadmap, SQL injection via unparameterized queries, OS command injection, default passwords, shipping with a component already on CISA's Known Exploited Vulnerabilities catalog, shipping open-source components with a CVSS 9.0+ vulnerability, insecure/deprecated cryptography such as TLS 1.0/1.1, MD5, SHA-1 or DES, and hardcoded credentials or secrets in source code); "Security features, which describe the security functionalities that a product supports" (2 items — no native or federated multi-factor authentication, and insufficient logging to reconstruct common intrusion types); and "Organizational processes and policies, which describe the actions taken by a software manufacturer to ensure strong transparency in its approach to security" (3 items — not issuing timely CVEs with a CWE field for critical/high-severity flaws, no published vulnerability disclosure policy, and not clearly communicating an on-premises product's support period). The document is explicit that this list "is focused and does not include every possible inadvisable cybersecurity practice" — omission from the list is not an endorsement.
What is the memory-safety roadmap requirement, and when is it due?
Bad practice #1 states that developing "new product lines for use in service of critical infrastructure or NCFs in a memory-unsafe language (e.g., C or C++) where readily available alternative memory-safe languages could be used is dangerous," and separately, that for existing products in memory-unsafe languages, "not having a published memory safety roadmap is dangerous." The recommended action sets a concrete deadline: "software manufacturers should publish a memory safety roadmap by the end of 2025, outlining their prioritized approach to eliminating memory safety vulnerabilities in priority code components written in memory unsafe languages" (e.g., network-facing code or cryptographic routines). This obligation is scoped — it "does not apply to products that have an announced end-of-support date that is prior to Jan. 1, 2030" — and the document's own Note acknowledges migration is incremental, citing interim steps like hardware/compiler mitigations and rewriting only the highest-risk components first. CWE-119 is the associated weakness class.
Is this guidance legally binding on device or software makers?
No. The document states this directly in its Overview: it is "voluntary guidance," and "this document imposes no requirement to do so" (to avoid the listed practices). It carries no enforcement mechanism of its own. That said, several of the same practices already show up as hard requirements elsewhere — default passwords and a vulnerability disclosure policy are PSTI and EU Cyber Resilience Act obligations, for example — so treating the catalog as voluntary-today, likely-regulated-tomorrow is a reasonable reading rather than dismissing it as non-actionable.
Sources
- CISA / FBI: Product Security Bad Practices, Version 2 (January 2025) — full text, all 13 bad practices, CWE mappings, Change Record
- CISA: Product Security Bad Practices (resource landing page)
- Federal Register: CISA Request for Information docket, PRODUCT SECURITY BAD PRACTICES (Docket No. CISA-2024-0028)
- CISA: Secure by Design (parent initiative)
- Federal Register: Request for Comment on Product Security Bad Practices Guidance (89 FR 83508, 16 October 2024)
- MITRE CWE-1392: Use of Default Credentials
- NIST SP 800-218, Secure Software Development Framework (SSDF) Version 1.1 (PDF)
- NIST SP 800-52 Rev. 2, Guidelines for the Selection, Configuration, and Use of TLS Implementations
- CISA Known Exploited Vulnerabilities catalog, JSON feed (cisagov/kev-data)
- CVE Program: CNA Operational Rules v4.0 (PDF)
- OWASP SQL Injection Prevention Cheat Sheet
- OWASP OS Command Injection Defense Cheat Sheet
- FIDO Alliance: Passkeys
- UK PSTI (Security Requirements for Relevant Connectable Products) Regulations 2023, Schedule 1
- Regulation (EU) 2024/2847 (Cyber Resilience Act), EUR-Lex
About this guide
SoftWSP Firmware Security Desk
Covers the Firmware vulnerabilities & fixes cluster: 131 guides.

In this cluster
Questions about this page?
