January 2025 Summaries

The release of DeepSeek R1, an open-source AI model from China, has garnered attention due to its potential as an alternative to models like ChatGPT and its impact on stock markets. While open-source models are generally seen as safer than proprietary ones, questions about the security and legal risks of using DeepSeek, particularly given its Chinese origin, are crucial for security teams. Endor Labs provides a methodology for evaluating these risks by analyzing models through a scoring system based on four risk categories: security, activity, popularity, and operational aspects. DeepSeek R1 has an overall Endor Score of 7 out of 10, with individual scores reflecting both positive aspects, such as its MIT license and customizable open-source weights, and concerns like the lack of a dataset and potential risks in example codes. Although the model itself is not inherently unsafe, users are advised to evaluate it within their threat models and organizational policies, especially given security vulnerabilities in its hosted service. The discourse around DeepSeek's open-source status and potential issues with model alignment and licensing highlights the need for ongoing scrutiny as organizations increasingly adopt AI models as critical dependencies in their software ecosystems.

Endor Labs aims to secure various dependencies in software applications, including open source AI models, by leveraging their capabilities to discover, evaluate, and enforce policies for using these models from platforms such as Hugging Face. As open source AI models introduce unique risks due to the combination of code, weights, and training data from multiple sources, Endor Labs provides tools to help application security teams manage these risks effectively. By using advanced program analysis, Endor Labs can inventory dependencies, including those not explicitly declared, and offers features to evaluate AI models based on security, activity, popularity, and quality. This functionality supports compliance with frameworks like ISO/IEC 42001:2023 and NIST-AI-600-1 by enabling comprehensive risk assessments and security control implementations. Additionally, Endor Labs provides tools for discovering AI models within source code and enforcing guardrails for their usage, while also offering policies to manage licensing risks. Their AI model discovery process has shown an 80% accuracy rate in identifying AI models within code, even when the model names are not directly specified, reflecting a satisfactory approach to enhancing application security in the evolving landscape of AI technologies.

In December 2024, the Apache Software Foundation disclosed a critical security vulnerability impacting Apache Struts, specifically a path-traversal bug in the FileUploadInterceptor class that can lead to remote code execution (RCE), posing a high risk for systems using versions prior to 6.4.0. RCE vulnerabilities are particularly dangerous as they allow attackers to take control of affected systems, potentially leading to data theft or further attacks. Not all users of Apache Struts, a widely used framework for building Java web applications, are affected; those at risk typically use the vulnerable file upload feature, though many may already have mitigations in place through server configurations or security plugins. Organizations must assess their risk level by identifying if they use the affected versions and features, and whether they have effective security measures in place. Addressing these vulnerabilities can be challenging, with major upgrades to safer versions potentially disrupting applications, especially if they involve significant changes. For those unable to upgrade immediately, Endor Patches offers a solution by providing backported security patches that remove high-severity vulnerabilities without causing application breaks, thus allowing organizations to maintain security while planning larger updates or sunsetting applications.

Opengrep emerges as a community-driven fork of Semgrep's static code analysis engine in response to Semgrep's decision to place critical features behind a commercial license, effectively restricting new community-contributed rules and essential tools for developers. By maintaining a truly open-source framework, Opengrep aims to provide a drop-in replacement that remains accessible and innovative, backed by a coalition of over 10 application security vendors such as Aikido Security, Orca Security, and others. This collaborative effort among competitors underscores the importance of a standardized open-source SAST engine, allowing developers and security teams to benefit from unrestricted access to scanning capabilities, community-driven feature development, and long-term stability through plans to transition to foundation oversight. Opengrep distinguishes itself from Semgrep by adhering to principles of true open-source access, community governance, and a commitment to foundation management, inviting global participation and contributions to ensure the evolution of security tools through collaboration rather than commercial interests.

Endor Labs offers customers a comprehensive analytics dashboard to enhance their vulnerability management programs by providing data-driven insights and evidence of remediation efforts. The dashboard includes features like the Vulnerabilities Snapshot, which provides a 90-day overview of newly discovered and resolved vulnerabilities, allowing users to filter by specific criteria such as severity or reachability. It also includes time-to-fix metrics, such as Mean Time to Remediate, to gauge the efficiency of addressing vulnerabilities. The dashboard's Vulnerabilities Over Time visualization enables teams to track trends and correlate changes with specific events, transforming raw data into compelling narratives for stakeholders. Additionally, the SLA compliance panel helps ensure no critical vulnerabilities are overlooked by tracking their proximity to breaching service level agreements. This tool aims to help customers tell a compelling security progress story, influence internal stakeholders, and meet regulatory requirements effectively.

FedRAMP compliance requires rigorous management of vulnerabilities, but false positives can complicate this process by inflating vulnerability counts without posing real threats. These false positives, which are vulnerabilities technically present but unexploitable within an application's specific context, need to be documented in a Plan of Action and Milestones (POA&M), potentially increasing the workload and cost of compliance. To address this, FedRAMP allows deviation requests, but justifying them can be challenging. Tools like Endor Labs utilize function-level reachability analysis to determine if a vulnerability is genuinely exploitable, thus identifying false positives with high confidence. This involves constructing detailed call graphs to map execution paths and verify if vulnerabilities can be exploited, allowing organizations to focus resources on actual threats. Endor Labs also aids in generating essential documentation, such as Software Bill of Materials (SBOM) and Vulnerability Exploitability Exchange (VEX), and provides continuous rescanning and alerting to maintain compliance and adapt to any changes in the risk profile.

Endor Labs addresses the challenges of securing open-source software dependencies by focusing on the most vulnerable libraries, which disproportionately contribute to security risks across the software supply chain. Research reveals that a small number of libraries account for the majority of vulnerabilities, with transitive dependencies often complicating upgrades due to potential breaking changes. By prioritizing the remediation of critical and high vulnerabilities found primarily in transitive dependencies, Endor Labs helps organizations like financial services firms significantly reduce their security risks, as demonstrated by one customer's ability to remediate 98.35% of such vulnerabilities using just nine Endor Patches. Through program analysis and upgrade impact assessment, Endor Labs offers solutions that not only identify vulnerabilities but also provide actionable remediation strategies, enabling developers to patch vulnerabilities without causing breaking changes and allowing upgrades to be tackled when teams are prepared.