March 2026 Summaries

San Francisco, a city known for its dynamic transformations, hosted the RSA Conference 2026, which was marked by a blend of enthusiasm and skepticism towards AI's pervasive role in cybersecurity and software development. With over 44,000 security professionals gathered, the conference highlighted the urgency of discerning genuine innovations from mere trends and emphasized the need for precision over flashy claims. A focal point was the Techstrong Seminar on AI-native development and security, where thought leaders discussed the importance of context and intent in AI-driven processes, advocating for a disciplined approach to managing AI agents as privileged insiders. Key discussions also centered on the evolving threat landscape, where trusted systems are increasingly exploited, and security teams must prioritize real-time validation over mere detection to keep pace with emerging challenges. The event underscored the necessity for robust governance frameworks to manage AI integration, with successful companies moving towards unified security narratives that address operational risks and foster trust. As the non-human identity and agentic AI space grows rapidly, the conference served as a platform for exploring its potential and defining its future trajectory, while also highlighting the need for security strategies that are both agile and precise.

Supply chain attacks by the PCP Team compromised key components such as AquaSecurity's trivy-action and the Python litellm package, affecting numerous repositories and packages, though the full extent remains challenging to determine. GitGuardian employed specific methodologies to assess the impact using available data, identifying 474 repositories that executed malicious trivy-action code and 1,705 PyPI packages susceptible to compromised litellm versions. These analyses, however, provide only a lower bound on the true impact, as private repositories and transitive dependencies complicate the full picture. The study emphasizes the importance of precise methodologies to evaluate potential vulnerabilities in organizations' environments, highlighting the broader risks posed by supply chain attacks, which can cascade through interconnected systems and affect high-profile companies.

BSides San Francisco 2026 highlighted the evolving challenges and strategies in cybersecurity, emphasizing the importance of adapting to rapid technological changes, particularly with the rise of AI. The event, which occurred just before the RSA Conference, attracted nearly 3,000 participants who engaged in numerous talks, workshops, and interactive sessions. Key discussions included Anna Westelius's analysis of security history as a series of pivots, emphasizing the transition from heroics to engineering, and Farshad Abasi's critique of traditional threat modeling, advocating for a dynamic feedback loop between models and real-world evidence. Bhaumik Shah addressed modern attacks on identity tokens, stressing the need for robust validation processes, while Vinod Tiwari highlighted the risks posed by malicious IDE extensions, urging teams to treat these as significant security concerns. The overarching theme was the necessity of integrating security into the core of operational design, focusing on identity, permissions, and trust management, to ensure resilience against evolving threats and to align security measures with the fast-paced development landscape.

The TeamPCP supply chain attack on LiteLLM packages 1.82.7 and 1.82.8 involved the use of infostealer malware that compromised SSH keys, cloud credentials, API tokens, and more from developer environments, creating significant security concerns for organizations using this popular AI development tool. Despite PyPI's quick removal of the infected packages, the attack's potential impact was extensive, as it involved critical infrastructure access keys. Organizations are urged to detect such compromises by monitoring for indicators of compromise and leveraging Endpoint Detection and Response (EDR) tools. The attack highlights the vulnerability of CI/CD pipelines and emphasizes the need for thorough audits of these systems to ensure no compromised packages were integrated. GitGuardian has developed a script to aid rapid detection of exposed secrets, allowing security teams to coordinate remediation efforts efficiently. The focus is on immediate incident response, prioritizing critical exposures, and integrating secret management systems to mitigate future risks. This incident underscores the importance of maintaining rigorous security practices in development environments, as threat actors increasingly target these areas where sensitive data often resides unmonitored.

The recent cybersecurity incidents involving Trivy and Shai Hulud highlight the evolving nature and impact of supply chain attacks. Trivy began as a targeted credential theft operation exploiting misconfigured workflows and incomplete remediation efforts, allowing attackers to compromise CI/CD pipelines, manipulate trusted GitHub Actions tags, and spread through Docker images and Kubernetes environments. Unlike Trivy, Shai Hulud was a broader, self-propagating attack with a focus on persistent and systemic disruption, using methods like backdooring npm packages and leveraging GitHub runners for command-and-control operations. Both incidents underscore the critical importance of comprehensive remediation, robust secrets management, and proactive security measures to prevent attackers from exploiting residual vulnerabilities. The lessons learned emphasize the need for swift detection, thorough credential rotation, and a deep understanding of potential blast radii to effectively contain breaches and prevent them from evolving into extended campaigns.

Supply chain security has increasingly focused on developer workstations as a crucial vulnerability point in the software lifecycle, as they are often the initial targets of modern attacks designed to harvest credentials and gain access to privileged systems. Developers' machines, where code is written and dependencies are managed, have become appealing targets for cyberattacks due to the sensitive information they contain, such as API keys, cloud tokens, and other credentials stored in plaintext. This shift in attack strategy highlights the importance of securing developer environments by eliminating plaintext secrets, adopting identity-based authentication, and using approved secret managers. While long-term security improvements are necessary, immediate measures like deploying honeytokens—decoy credentials that alert defenders of unauthorized access—can help detect and mitigate breaches early. Organizations must support developers in these efforts by providing clear security guidelines and tools, ensuring that both individual actions and broader enterprise policies align to protect against potential threats.

Chainguard Assemble 2026, an event held at The Glass House, gathered approximately 400 security-minded professionals to discuss the transformation of software systems into reliable and governed frameworks amid rapidly evolving AI-powered threats. Key discussions highlighted the necessity of embedding security within the design of automated systems, rather than retrofitting it, as traditional "scan and patch" methods are insufficient in the face of swift dependency changes and automation. Dan Lorenc's keynote emphasized the parallels between AI in software development and power tools in woodworking, advocating for controlled dependencies and transparent processes. Sessions by Molly Soja and Ayesha Bhutto discussed the importance of golden images in maintaining consistency and compliance, while Brandon Heard's talk focused on enhancing developer productivity without compromising security. Collin Estes highlighted NASA's approach to automated compliance in mission-critical environments, showcasing the shift from brand trust to process trust and the growing importance of managing non-human identities within governance models. The overarching message was that security must adapt to faster development cycles by constraining system inputs and ensuring trust through verifiable processes, as AI and automation redefine the landscape of software production.

In 2025, the rapid integration of AI into software development significantly accelerated the pace of coding and broadened participation, but it also led to a substantial increase in security vulnerabilities, particularly with the rise in hardcoded secrets in public and internal code repositories. The "State of Secrets Sprawl" report highlights a 34% year-over-year increase in hardcoded secrets on public GitHub, with AI services increasingly contributing to these leaks, as seen with the 81% rise in AI service secrets. The report underscores the complexity of managing secrets in a rapidly evolving ecosystem where AI tools, APIs, and service accounts multiply the surface area for potential attacks. This issue is compounded by unsafe practices encouraged by official documentation and the normalization of hardcoded credentials, resulting in significant security debts in private repositories. The problem extends beyond code repositories to collaboration tools and local environments, where secrets are often shared under urgent circumstances, leading to high-impact exposures. The persistence of valid but unremediated credentials from previous years further complicates the security landscape, emphasizing the need for improved governance of non-human identities (NHI) to keep pace with AI adoption. The report advocates for a shift from mere detection to comprehensive prevention and lifecycle management strategies that address the nuances of AI-driven software environments.

In the realm of cybersecurity and platform engineering, a significant challenge arises from the lack of accountability for machine identities, such as service accounts and API keys, which often outnumber human identities but lack clear ownership. This absence of accountability results in time-consuming and potentially risky scenarios when compromised credentials are detected, as security teams scramble to identify responsible parties. GitGuardian addresses this issue with NHI Governance by automatically suggesting and managing owners for non-human identities, leveraging existing tools and data to propose candidates based on strong signals. This system allows for flexible assignment and management of ownership, ensuring that every identity has a designated owner, thereby streamlining incident response, improving security hygiene, and facilitating compliance with audit requirements. By implementing ownership, organizations can shift focus from reactive credential management to proactive security measures, ultimately enhancing organizational security and efficiency.

Montreal's urban policy, which restricts buildings from surpassing the height of Mount Royal, serves as a metaphor for the ConFoo 2026 conference's focus on balancing growth with protective measures. The conference, attended by around 800 participants, spanned five days and featured over 190 sessions that emphasized the importance of implementing guardrails in fast-moving technology systems to prevent unexpected outcomes. Key discussions highlighted the need for modern security measures, such as Zero Trust models to authenticate non-human agents, robust prompt hygiene for language models to prevent adversarial manipulation, and comprehensive dependency management to mitigate supply chain attacks. Speakers stressed the importance of treating AI and software systems with a structured approach, emphasizing context-based access, continuous evaluation, and the adoption of operational disciplines to enhance security and resilience. The overarching theme was the necessity for adaptable yet controlled growth in technology, mirroring Montreal's approach to urban development.

In 2026, non-human identities (NHIs), such as service accounts, API keys, and OAuth applications, have become a significant security concern as they now outnumber human users in most cloud-native organizations. These NHIs present vulnerabilities due to unmanaged lifecycles, overprivileged access, and exposed credentials, which can be exploited by attackers who focus on the sprawling web of machine identities rather than traditional password hacking. Security strategies must therefore evolve to address this by implementing multi-layered approaches that encompass detection, lifecycle management, governance, and compliance to effectively reduce risks and prevent data breaches. The rapid growth of NHIs, fueled by automation and cloud-native architectures, necessitates robust security tools that provide comprehensive visibility, automate identity management, and enforce least-privilege principles across infrastructures. As the landscape evolves, organizations must integrate advanced NHI security solutions with existing IAM and vault systems, focusing on proactive lifecycle automation and continuous monitoring to secure both current and future infrastructures effectively.

In collaboration with Google, GitGuardian conducted an extensive analysis of private key leaks, highlighting significant security vulnerabilities in the TLS ecosystem, with a focus on the difficulty of assessing the real-world impact when private keys leak on platforms like GitHub and DockerHub. The research uncovered that, since 2021, nearly one million unique private keys were leaked, with over 40,000 mapped to real TLS certificates, including those protecting major organizations. Despite notifying organizations about their exposed certificates, the response rate was alarmingly low, revealing a widespread misunderstanding of private key risks. The study underscores systemic challenges, such as the inadequate use of revocation mechanisms and the persistence of compromised keys, which often outlive certificate renewals. It calls for industry-wide changes, including shorter cryptoperiods, mandatory key rotation, and the prohibition of private key reuse, to mitigate these vulnerabilities. The comprehensive findings, presented at the Real World Crypto 2026 conference, stress the urgent need for technical innovation and cross-organizational partnerships to protect internet security at scale.

Enterprise security traditionally focuses on data center and cloud infrastructure protection, but the developer workstation is increasingly a critical entry point for supply chain attacks, as it often contains locally stored credentials that attackers can exploit. Developers, needing access to internal systems, are now targets due to their creation and management of various credentials, which are often stored in plaintext and can be easily harvested by attackers. Tools like ggshield from GitGuardian are recommended to scan and detect secrets across developer environments, while best practices suggest moving credentials into secure vaults or password managers, using SOPS for encrypting necessary .env files, and employing global .gitignore to prevent accidental commits of sensitive data. Additionally, adopting authentication mechanisms like WebAuthn and OIDC can eliminate the need for storing long-lived secrets, while ephemeral credentials and identity-based authentication workflows, such as those using SPIFFE, help reduce the risk of credential theft. The shift towards securing developer workstations is crucial for minimizing the overall risk of unauthorized access within enterprise systems.