June 2026 Summaries

Black Hat USA, set in Las Vegas from August 1-6, 2026, is a key event in the security industry known for its rich concentration of security knowledge, innovative research, and networking opportunities. It kicks off the renowned "Hacker Summer Camp," which also includes DEF CON and BSidesLV, alongside various ancillary events. Attendees are advised to prioritize original research presentations over marketing pitches, with standout talks by experts like Christopher Domas and James Kettle focusing on groundbreaking discoveries in security. Arsenal, a showcase for open-source security tools, provides direct interaction with developers, offering insights into emerging technologies. The conference also features specialized Summits on topics such as AI and financial threats, while the Business Hall offers strategic opportunities to engage with engineers and explore industry trends. The event emphasizes the importance of hallway conversations for information exchange and advises participants to pace themselves for the marathon of activities extending into DEF CON, underscoring the value of networking and learning over collecting vendor swag.

The software development landscape is undergoing significant changes as AI agents enable both traditional engineers and citizen developers to contribute to production code, leading to a surge in unreviewed code and increased software vulnerabilities. The industry is grappling with the challenge of timely vulnerability detection and resolution, as traditional methods like human review and post-code writing tools fail to keep up with the pace of development. Semgrep Guardian offers a solution by providing an agentic code security platform that operates within Integrated Development Environments (IDEs) to automatically scan and fix AI-generated code as it is written, ensuring real-time security and preventing critical vulnerabilities such as OWASP Top 10 issues, malicious packages, and hardcoded secrets. Guardian boasts partnerships with major platforms like GitHub Copilot and offers easy deployment across development teams, allowing security teams to maintain full visibility and control over code integrity without disrupting workflow. The platform's efficiency is demonstrated by its ability to perform millions of quick scans weekly, preventing costly vulnerabilities from being introduced into production environments.

An experiment conducted by Semgrep evaluated various open-source models against their IDOR benchmark to assess vulnerability-detection performance, revealing unexpected results. Among the models tested, Zhipu AI's GLM 5.2, an open-weight model, achieved a 39% F1 score, outperforming Claude Code's 32% score at a significantly lower cost of approximately $0.17 per vulnerability detected. However, it still lagged behind Semgrep's multimodal pipeline, which achieved 53–61% F1 scores with a more sophisticated harness. The test primarily aimed to discern how much of the performance was attributable to the model itself versus the harness—a critical question for security tasks leveraging AI. GLM 5.2, notable for its open-weight nature and cost-effectiveness, showed promise despite not having endpoint discovery support like the multimodal pipeline, indicating that open-weight models have become a viable consideration for security research. The experiment underscores the importance of harness configuration and represents a step forward in the competitiveness of open-weight models, although it also highlights that one successful outcome does not imply universal superiority across different tasks or datasets.

The text discusses the application of AI-powered detection tools, specifically Semgrep Multimodal, in identifying code vulnerabilities compared to other models such as Opus 4.8 and GPT 5.5. It highlights that while AI models can be effective at identifying some vulnerabilities, they often miss a significant portion of the code due to a lack of comprehensive coverage. Semgrep Multimodal, however, employs deterministic program analysis to ensure thorough examination of codebases, leading to significantly higher recall rates and reduced cost per true positive compared to other approaches. The text emphasizes the importance of both the AI model and the surrounding architecture or scaffolding, suggesting that combining improved models with structured workflows yields better overall performance. It concludes that effective risk reduction in AI-powered security tools depends on the balance between model capabilities and the orchestrating framework around them.

npm, the default package manager for Node.js and one of the largest software registries, has long faced criticism for its automatic execution of lifecycle scripts during package installations, which posed significant security risks. This feature allowed arbitrary shell scripts from third-party code to run without user consent, a vulnerability exploited in numerous supply chain attacks. These attacks often involved compromised packages delivering malware via postinstall scripts, even affecting packages buried deep in dependency trees. In response to these security issues, npm v12 will default to disabling these scripts, requiring explicit user permission to run them, aligning with the practices of other package managers like pnpm, Bun, and Yarn. While legitimate use cases for lifecycle scripts existed, such as compiling native addons or downloading binaries, the default trust model left systems exposed to attacks like the infamous Axios incident. Despite years of calls for change, npm's shift to a safer default setting marks a significant step in securing the JavaScript ecosystem, though other supply chain threats persist.

Semgrep's file targeting step, crucial for filtering files against ignore patterns before scanning, experienced significant inefficiencies, taking hours in some cases due to millions of regex calls for large repositories. By replacing most regex lookups with string comparisons and building a hash table index, Semgrep drastically improved performance, reducing a customer's repo scan time from 7.5 hours to under 2 minutes. These changes, available in Semgrep 1.162.0, reduced the 99th percentile scan duration from nearly an hour to under 12 minutes. Semgrep supports various ignore patterns, including those from .gitignore and .semgrepignore, allowing customization of scan findings and optimizing scan times further by focusing on relevant files. The implementation of optimized matching strategies, inspired by Ripgrep, enabled most patterns to be evaluated with simple string comparisons, significantly decreasing regex calls. The improvements led to substantial speedups across Semgrep's customer base, especially for the most time-consuming scans, enhancing the overall efficiency and reliability of Semgrep's scanning process.

Semgrep Pro Engine 1.158.0 introduces a redesigned taint analysis engine that significantly enhances performance, resulting in up to 75% speed improvements on full scans. This update is part of a larger effort to improve Semgrep's efficiency, which began with the development of a continuous profiler called Pyro Caml for its OCaml-based core. The redesign focused on eliminating the need to run taint analysis twice, a practice that previously consumed a substantial portion of CPU time and hindered parallelization. By integrating OCaml Multicore, the team was able to refactor the code to perform taint analysis once, enabling parallel processing and reducing CPU and memory usage. This change has led to a notable reduction in scan times, with the 95th percentile decreasing from 10 minutes to 7 minutes and 30 seconds, and the 99th percentile from 45 to 35 minutes. The improvements are particularly impactful for large repositories, where scan times have decreased significantly, fostering greater consistency and reliability in Semgrep's performance.

Miasma, a self-spreading npm worm, has compromised 57 npm packages through over 286 malicious versions, using an unconventional method involving a 157-byte binding.gyp file for code execution during npm install without lifecycle scripts. This malware rapidly spreads across npm packages, harvesting sensitive credentials from AWS, GCP, Azure, and GitHub Actions secrets, and exfiltrating them to attacker-controlled GitHub repositories. It further injects persistent backdoors into AI coding-assistant configuration files, poisoning AI-generated code and mimicking Mini Shai-Hulud by forging provenance attestations to appear legitimate. The attack necessitates urgent scans and credential rotations for affected systems, with indicators of compromise including specific exfiltration path patterns and C2 beacon keywords.

Semgrep's core SAST engine is built in OCaml, a language with a limited ecosystem that lacks essential libraries for observability, crucial for its operations on numerous code repositories. To bridge this gap, the Semgrep team developed Pyro Caml, a continuous profiler tailored for OCaml, which they recently released as version 1.0.0. Continuous profiling is distinct from traditional profiling as it operates in production, constantly analyzing and reporting program performance data to a centralized system. This approach is vital for Semgrep, especially given its security constraints and the need to avoid handling user code directly. Existing profilers like ocamlprof and tools dependent on Linux's perf_event_open were incompatible due to Semgrep's use of gVisor, which led to the creation of Pyro Caml using the Pyroscope SDK. Pyro Caml leverages OCaml's Memprof for call stack sampling, ensuring low overhead while providing insights into bottlenecks within the codebase. The profiler's development involved overcoming challenges related to sampling accuracy and runtime impact, and it is now a critical tool in Semgrep's performance optimization efforts, having been deployed across millions of scans. Despite some limitations, such as missing samples in certain contexts and challenges with recursive calls, Pyro Caml has proven effective, and future enhancements are anticipated to address these issues and expand its capabilities.

RedHat recently fell victim to an npm worm attack that compromised numerous packages in the @redhat-cloud-services npm organization, replacing the Dune theme with Greek Mythology elements, and is suspected to be a variant of the Mini Shai-Hulud malware. This attack utilizes multi-stage credential harvesters that activate through preinstall hooks during npm installations, targeting secrets from various platforms including GitHub Actions, AWS, GCP, Azure, Kubernetes, and more. The breach was facilitated by compromised GitHub Actions OIDC tokens, and the malicious packages were distributed through the RedHatInsights/javascript-clients repository. The worm, which modifies its approach to seek more cloud environment credentials, exfiltrates data by reading directly from GitHub Actions Runner’s memory and collecting various credentials, such as AWS and Azure tokens, npm publish tokens, and .env files, potentially impacting CI/CD environments significantly.