It’s a Paper Trail for the week ending in March 28, 2026, and we’ll cover what happened last week in the Information Security space.
Last Week, in Review
- Google’s Mandiant M-Trends 2026 report – based on 500,000+ hours of incident response in 2025 – identified that hand-off time between initial access brokers and follow-on adversaries has collapsed to a median of 22 seconds from 8+ hours in 2022, with global median dwell time rising from 11 days to 14 days. Furthermore, ransomware attacks are not only leading to an organization’s data being encrypted but also to destruction of backups, identity services and virtualization management planes to deny recovery. [Google]
- The introduction of Semgrep Multimodal and GitHub’s AI-powered bug detection alongside CodeQL points to a growing consensus that pure static analysis and pure AI based security code scanning fall short on their own. As AI supercharges both code output and exploit development, the industry is converging on the defense of merging them together to help keep pace with deterministic precision, and amplified machine reasoning. [GitHub, Semgrep]
Pulse
Deepfake technology is now cheap but realistic enough that adversaries can use AI-cloned voices and video to impersonate executives and authorize fraudulent wire transfers – creating a possibility of incidents going unreported because employees believe that they are talking to real colleagues. No U.S. law requires companies to disclose a successful deepfake attack, leaving a significant regulatory blind spot as the threat scales. Organizations should combine simulated deepfake drills, AI detection capability, and – most critically – mandatory out-of-band verification for any digital request involving payments or system access. [Bloomberg Law]
A sophisticated multi-stage, multi-ecosystem supply chain compromise was used to gain a foothold and compromise LiteLLM – a package used by AI frameworks relied upon by Stripe, Netflix, Google, and thousands of startups – and found in 36% of all cloud environments according to Wiz, meaning any developer or pipeline that installed the package unknowingly pulled down credential-stealing malware alongside it. The attack is notable not just for its scale, but its method where adversaries first compromised Aqua Security’s Trivy – a widely trusted security scanning tool – to gain a foothold in the software supply chain, then used that position to push malicious code into LiteLLM on PyPI. Since malware arrived through a trusted and routine update, developers and automated pipelines had no reason to suspect it. Organizations should audit any Python environments, CI/CD pipelines, or containers that installed LiteLLM between 10:39 – 16:00 UTC on March 24 for compromised versions 1.82.7 and 1.82.8 of LiteLLM. If these versions are found, all secrets from affected systems should be rotated. [DataDog, Snyk, The Record]
Fix-it Frank
Insufficient input validation leading to memory overread vulnerability (CVE-2026-3055) in Citrix NetScaler ADC and NetScaler Gateway versions prior to 13.1-62.23 for 13.x series or 14.1-66.59 for 14.x series are under active reconnaissance and – if exploited – could allow an adversary to read sensitive contents from the memory. It is recommended to upgrade these appliances to the latest and supported versions to remediate vulnerabilities. [Citrix, watchTowr Labs, The Hacker News]
Apple issues emergency software updates for older versions of iOS and iPadOS after Coruna – an exploit kit chaining multiple vulnerabilities to compromise the device OS versions 13 to 17.2.1 – was identified as being actively used. These patches will be available for iPhone 6s, iPhone 7, iPhone SE (1st gen), iPhone 8, iPhone 8 Plus, and iPhone X , iPad Air 2, iPad mini (4th gen). Organizations still running these older devices with older software versions should immediately rollout upgrade, or at the least retire the devices if they are no longer supported. [SecurityWeek, Google]
PolyShell – a vulnerability in Magento’s REST API’s file upload functionality in cart item custom options stems from insufficient validation against polyglot files, which can act as both an image and a script and depending on server configuration, this can lead to remote code execution or account takeover through stored cross-site scripting (XSS). While the complete remediation for this vulnerabilities is not yet available, mitigating factors such as 3rd party patch (markshust/magento-polyshell-patch) should be applied along with configuring nginx/apache to block access to files within pub/media/custom_options directory. [Bleeping Computer, Searchlight Cyber]
The Fine Print
On March 20, 2026, the White House released a policy framework for AI, recommending congress to promote development of Artificial Intelligence, establish a unified federal standard for AI regulations in place of state-by-state approach, and preserve state authority over child safety and consumer protection. [Davis Wright Tremaine, Holland & Knight]
Oklahoma signed SB 546 into law on March 20, 2026, making it the 21st state with a comprehensive consumer data privacy law, effective January 1, 2027. Modeled after Virginia and Tennessee’s frameworks, the law applies to organizations processing personal data of 100,000 or more Oklahoma residents – or 25,000 or more if over 50% of revenue comes from data sales – and includes consumer rights around access, correction, deletion, and opt-out, along with a mandatory 30-day right to cure. [Davis Wright Tremaine, Hunton Andrews Kurth]
The Bottom Line
The perimeter is no longer a place defenders can defend from. When the tools built to protect pipelines become the delivery mechanism for malware, when handoffs between threat actors happen faster than an alert can fire, and when a single deepfake call can authorize a wire transfer, the defenses of the past demand adaptation – hybrid detection, out-of-band verification, and improved supply chain rigor are no longer optional layers, they are the baseline.
—
Thanks for tuning-in to this edition of Paper Trail. We’d love your ideas and suggestions, so email us at feedback@hackwithheart.com. You can follow Paper Trail wherever you like best – read it on hackwithheart.com, listen on Spotify or Apple Podcasts, or watch on YouTube.