Perform code review for high-risk applications in an opportunistic fashion. For example, organizations can follow up a design review with a code review looking for security issues in source code and dependencies and perhaps also in deployment artifact configuration (e.g., containers) and automation metadata (e.g., infrastructure-as-code). This informal targeting often evolves into a systematic approach (see [CR1.4]). Manual code review could be augmented with the use of specific tools and services, but it has to be part of a proactive process. When new technologies pop up, new approaches to code review might become necessary.

Incorporate static analysis into the code review process to make the review more efficient and consistent. Automation won’t replace human judgment, but it does bring definition to the review process and security expertise to reviewers who typically aren’t security experts. Note that a specific tool might not cover an entire portfolio, especially when new languages are involved, so additional local effort might be useful. Some organizations might progress to automating tool use by instrumenting static analysis into source code management workflows (e.g., pull requests) and delivery pipeline workflows (build, package, and deploy) to make the review more efficient, consistent, and aligned with release cadence. Whether use of automated tools is to review a portion of the source code incrementally, such as a developer committing new code or small changes, or to conduct full analysis by scanning the entire codebase, this service should be explicitly connected to a larger SSDL defect management process applied during software development. This effort is not useful when done just to “check the security box” on the path to deployment.

A security-focused code review is mandatory for all software projects, with a lack of code review or unacceptable results stopping a release, slowing it down, or causing it to be recalled. While all projects must undergo code review, the process might be different for different kinds of projects. The review for low-risk projects might rely more heavily on automation (see [CR1.4]), for example, whereas high-risk projects might have no upper bound on the amount of time spent by reviewers. Having a minimum acceptable standard forces projects that don’t pass to be fixed and reevaluated. A code review tool with nearly all the rules turned off (so it can run at CI/ CD automation speeds, for example) won’t provide sufficient defect coverage. Similarly, peer code review or tools focused on quality and style won’t provide useful security results.

Mentors show developers how to get the most out of code review tools, including configuration, triage, and remediation. Security champions, DevOps and site reliability engineers, and SSG members often make good mentors. Mentors could use office hours or other outreach to help developers establish the right configuration and get started on interpreting and remediating results. Alternatively, mentors might work with a development team for the duration of the first review they perform. Centralized use of a tool can be distributed into the development organization or toolchains over time through the use of tool mentors, but providing installation instructions and URLs to centralized tool downloads isn’t the same as mentoring. Increasingly, mentorship extends to code review tools associated with deployment artifacts (e.g., container security) and infrastructure (e.g., cloud configuration). While AI is becoming useful to augment human code review guidance, it likely doesn’t have the context necessary to replace it.

Create and use custom rules in code review tools to help uncover security defects specific to the organization’s coding standards or to the framework-based or cloud-provided middleware the organization uses. The same group that provides tool mentoring (see [CR1.7]) will likely spearhead this customization. Custom rules are often explicitly tied to proper usage of technology stacks in a positive sense and avoidance of errors commonly encountered in a firm’s codebase in a negative sense. Custom rules are also an easy way to check for adherence to coding standards (see [CR3.5]). To reduce the workload for everyone, many organizations also create rules to remove repeated false positives and to turn off checks that aren’t relevant.

Maintain a living list of the most important kinds of bugs the organization wants to eliminate from its code and use it to drive change. Many organizations start with a generic list pulled from public sources, but broad-based lists such as the OWASP Top 10 rarely reflect an organization’s bug priorities. Build a valuable list by using real data gathered from code review (see [CR2.8]), testing (see [PT1.2]), software composition analysis (see [SE3.8]), and actual incidents (see [CMVM1.1]), then prioritize it for prevention efforts. Simply sorting the day’s bug data by number of occurrences won’t produce a satisfactory list because the data changes so often. To increase interest, the SSG can periodically publish a “most wanted” report after updating the list. One potential pitfall with a top N list is that it tends to include only known problems. Of course, just building the list won’t accomplish anything—everyone has to use it to find and fix bugs.

The defects found during code review are tracked in a centralized repository that makes it possible to do both summary and trend reporting for the organization. Reported defects drive engineering improvements such as enhancing processes, updating standards, adopting reusable frameworks, etc. For example, code review information is usually incorporated into a CISO-level dashboard that can include feeds from other security testing efforts (e.g., penetration testing, composition analysis, threat modeling). Given the historical code review data, the SSG can also use the reports to demonstrate progress (see [SM3.3]) or drive the training curriculum. Individual bugs make excellent training examples (see [T2.8]). Some organizations have moved toward analyzing this data and using the results to drive automation (see [ST3.6]).

Combine application security testing (AST) results so that multiple testing techniques feed into one reporting and remediation process. In addition to code review, testing techniques often include dynamic analysis, software composition analysis, container scanning, cloud services configuration review, etc. The SSG might write scripts or acquire software to gather data automatically and combine the results into a format that can be consumed by a single downstream review and reporting solution. The tricky part of this activity is normalizing vulnerability information from disparate sources

that might use conflicting terminology or scoring. In some cases, using a standardized taxonomy (e.g., a CWE-like approach) can help with normalization. Combining multiple sources helps drive better-informed risk mitigation decisions and identify engineering improvements.

When a security bug is found during code review (see [CR1.2], [CR1.4]), the organization searches for and then fixes all occurrences of the bug, not just the instance originally discovered. Searching with custom rules (see [CR2.6]) makes it possible to eradicate the specific bug entirely without waiting for every project to reach the code review portion of its lifecycle. This doesn’t mean finding every instance of every kind of cross-site scripting bug when a specific example is found—it means going after that specific example everywhere. A firm with only a handful of software applications built on a single technology stack will have an easier time with this activity than firms with many large applications built on a diverse set of technology stacks. A new development framework or library, rules in RASP or

a next-generation firewall, or cloud configuration tools that provide guardrails can often help in (but not replace) eradication efforts.

Use automated code review to identify malicious code written by in-house developers or outsource providers. Examples of malicious code include backdoors, logic bombs, time bombs, nefarious communication channels, obfuscated program logic, and dynamic code injection. Although out-of-the-box automation might identify some generic malicious-looking constructs, custom rules for the static analysis tools used to codify acceptable and unacceptable patterns in the organization’s codebase will likely become a necessity. Manual review for malicious code is a good start but insufficient to complete this activity at scale. While not all backdoors or similar code were meant to be malicious when they were written (e.g.,

a developer’s feature to bypass authentication during testing), such things tend to stay in deployed code and should be treated as malicious until proven otherwise. Discovering some types of malicious code will require dynamic testing techniques.

A violation of secure coding standards is sufficient grounds for rejecting a piece of code. This rejection can take one or more forms, such as denying a pull request, breaking a build, failing quality assurance, removing from production, or moving the code into a different development workstream where repairs or exceptions can be worked out. The enforced portions of an organization’s secure coding standards (see [SR3.3]) often start out as a simple list of banned functions or required frameworks. Code review against standards must be objective—it shouldn’t become a debate about whether the noncompliant code is exploitable. In some cases, coding standards are specific to language constructs and enforced with tools (e.g., codified into SAST rules). In other cases, published coding standards are specific to technology stacks and enforced during the code review process or by using automation. Standards can be positive (“do it this way”) or negative (“do not use this API”), but they must be enforced.