Industry shifts, security alerts, hardware, and AI/ML Compiled for immediate developer deployment.
calendar_todaysummarizeWeek 13-2026bolt2 CRITICAL
article
CanisterWorm: How a Self-Propagating npm Worm Is Spreading Backdoors Across the Ecosystem
TAG: SECURITY
StepSecurity researchers have identified CanisterWorm, a self-propagating npm worm deployed by the TeamPCP threat actor as a direct continuation of the Trivy v0.69.4 supply-chain compromise. The worm uses npm tokens stolen from CI/CD pipelines running the backdoored Trivy actions to autonomously publish malicious patch versions to every accessible npm namespace, including 16+ packages in the @opengov scope. Each compromised package version installs a persistent Python backdoor via a postinstall hook, establishes a systemd user service for persistence without requiring root, and polls a command-and-control endpoint hosted on the Internet Computer Protocol (ICP) blockchain at tdtqy-oyaaa-aaaae-af2dq-cai.raw.icp0.io, making the C2 highly resistant to takedown. A second-stage payload delivers destructive Kubernetes capabilities including a privileged DaemonSet named host-provisioner-iran targeting every node in a cluster. Defenders should immediately check for ~/.local/share/pgmon/service.py and rotate all npm tokens and CI/CD secrets on any machine that installed affected packages.
Trivy Compromised a Second Time: Malicious v0.69.4 Release and GitHub Actions Backdoored
TAG: SECURITY
On March 19, 2026, the aquasecurity/trivy-action GitHub Action was compromised for approximately 12 hours, with all tags from 0.0.1 through 0.34.2 pointing to a malicious commit containing a credential stealer. The injected code harvested GitHub Actions Runner worker memory via /proc/pid/mem, encrypted stolen secrets with an RSA-4096 public key, and exfiltrated them to the typosquatted domain scan.aquasecurtiy.org (resolving to 45.148.10.212 in Amsterdam). The aquasecurity/setup-trivy action and the trivy binary release v0.69.4 were similarly backdoored. This was the second compromise in three weeks, following the hackerbot-claw incident on February 28. StepSecurity Harden-Runner detected the attack in real-time across 12,000+ monitored public repositories, and analysis confirmed 45 repositories executed the compromised action, with 5 having custom secrets including AWS credentials and Docker Hub tokens directly accessible. Homebrew executed an emergency downgrade to v0.69.3. Affected teams should rotate GITHUB_TOKEN and all explicitly referenced secrets immediately.
Updates to GitHub Copilot Interaction Data Usage Policy
TAG: AI POLICY
Starting April 24, GitHub will use interaction data from Copilot Free, Pro, and Pro+ users to train and improve its AI models, unless users opt out via the Privacy settings page. The policy covers inputs, outputs, accepted or modified code suggestions, cursor context, file names, repository structure, and feedback ratings, and data may be shared with GitHub affiliates including Microsoft. Copilot Business and Enterprise users are not affected. GitHub says the decision is driven by measurable quality gains already observed from training on Microsoft employee interaction data, including higher suggestion acceptance rates across multiple programming languages. Users who previously opted out retain that preference automatically and will not be enrolled unless they actively opt in.
A Year of Open Source Vulnerability Trends: CVEs, Advisories, and Malware
TAG: SECURITY
GitHub published 4,101 reviewed advisories in 2025, the fewest since 2021, but this reflects a backlog depletion rather than improved security — newly reported vulnerability reviews actually rose 19% year over year. npm malware advisories surged 69% compared to 2024, driven largely by campaigns such as SHA1-Hulud, marking the highest count since malware tracking began in 2022. CWE-79 (cross-site scripting) remained the top weakness, while resource exhaustion (CWE-400, CWE-770), unsafe deserialization (CWE-502), and server-side request forgery (CWE-918) were unusually prevalent. Advisories missing any CWE classification dropped 85%, from 452 in 2024 to 65 in 2025. GitHub's CNA published 35% more CVE records than in 2024, outpacing the overall CVE Project's 21% growth, and 987 organizations submitted CVEs during the year.
Mozilla and Mila Announce Strategic Research Partnership to Advance Open Source and Sovereign AI
TAG: OPEN SOURCE AI
Mozilla and Mila, the Quebec Artificial Intelligence Institute, have announced a strategic research partnership — Mozilla's first-ever collaboration with a major AI research lab. The inaugural project focuses on trust and usability, including private memory architectures for AI agents, with the goal of demonstrating that open-source AI can match cutting-edge research capabilities. Mozilla President Mark Surman and Mila CEO Valerie Pisano both emphasized Canada's potential to lead on sovereign AI, citing the country's research depth and commitment to trustworthy, transparent systems. The partnership aims to reduce dependence on closed AI systems by developing technologies that enable transparency, accountability, and shared innovation across geographies, welcoming participation from research institutions and developers worldwide.
The Next.js Adapter API Just Shipped: What's Next for Multi-Platform Deployment
TAG: FRAMEWORKS
The Next.js Deployment Adapter API reached stable status in Next.js 16.2, a milestone achieved after 18 months of collaboration between Netlify, Cloudflare, Google/Firebase, AWS Amplify, OpenNext, and the Next.js team. Vercel's open-source adapter is already powering some production builds on their platform, while Netlify's and Cloudflare's adapters are in active development. The Next.js team has published the full 9,000+ end-to-end test suite as a conformance contract and established the Next.js Ecosystem Working Group with public meeting notes. Netlify's post also candidly details three architectural challenges the adapter API does not yet solve: Partial Prerendering (PPR) requires bespoke CDN infrastructure to stitch static and dynamic streams; cache revalidation demands atomic updates of coupled cache entries with no standard HTTP mechanism available; and on-demand revalidation is fundamentally incompatible with immutable CDN deploy models used by Netlify, Cloudflare Pages, and others.
Two AI Models Set to Stir Government Urgency, But Will This Challenge Undo Them?
This AI Explained video covers reports that OpenAI's upcoming Spud model and Anthropic's next Claude release are both expected to represent major capability leaps, with Anthropic reportedly warning US government officials that the new Claude series will significantly supercharge offensive and defensive cyber capabilities. OpenAI has shut down the Sora app and shelved its erotic chatbot to redirect compute toward Spud, while the Pentagon may be reconsidering its breach with Anthropic. The video also provides a thorough analysis of ARC-AGI-3, a new interactive benchmark in which humans score 100% but current frontier models including Gemini 3.1 score below 0.5%, with the low scores attributed to the benchmark's requirements for multi-step planning, in-context memory, and adaptive goal-setting across game-like environments rather than static pattern recognition.
CanisterWorm: How a Self-Propagating npm Worm Is Spreading Backdoors Across the Ecosystem
StepSecurity researchers have identified CanisterWorm, a self-propagating npm worm deployed by the TeamPCP threat actor as a direct continuation of the Trivy v0.69.4 supply-chain compromise. The worm uses npm tokens stolen from CI/CD pipelines running the backdoored Trivy actions to autonomously publish malicious patch versions to every accessible npm namespace, including 16+ packages in the @opengov scope. Each compromised package version installs a persistent Python backdoor via a postinstall hook, establishes a systemd user service for persistence without requiring root, and polls a command-and-control endpoint hosted on the Internet Computer Protocol (ICP) blockchain at tdtqy-oyaaa-aaaae-af2dq-cai.raw.icp0.io, making the C2 highly resistant to takedown. A second-stage payload delivers destructive Kubernetes capabilities including a privileged DaemonSet named host-provisioner-iran targeting every node in a cluster. Defenders should immediately check for ~/.local/share/pgmon/service.py and rotate all npm tokens and CI/CD secrets on any machine that installed affected packages.
Two AI Models Set to Stir Government Urgency, But Will This Challenge Undo Them?
This AI Explained video covers reports that OpenAI's upcoming Spud model and Anthropic's next Claude release are both expected to represent major capability leaps, with Anthropic reportedly warning US government officials that the new Claude series will significantly supercharge offensive and defensive cyber capabilities. OpenAI has shut down the Sora app and shelved its erotic chatbot to redirect compute toward Spud, while the Pentagon may be reconsidering its breach with Anthropic. The video also provides a thorough analysis of ARC-AGI-3, a new interactive benchmark in which humans score 100% but current frontier models including Gemini 3.1 score below 0.5%, with the low scores attributed to the benchmark's requirements for multi-step planning, in-context memory, and adaptive goal-setting across game-like environments rather than static pattern recognition.
Trivy Compromised a Second Time: Malicious v0.69.4 Release and GitHub Actions Backdoored
On March 19, 2026, the aquasecurity/trivy-action GitHub Action was compromised for approximately 12 hours, with all tags from 0.0.1 through 0.34.2 pointing to a malicious commit containing a credential stealer. The injected code harvested GitHub Actions Runner worker memory via /proc/pid/mem, encrypted stolen secrets with an RSA-4096 public key, and exfiltrated them to the typosquatted domain scan.aquasecurtiy.org (resolving to 45.148.10.212 in Amsterdam). The aquasecurity/setup-trivy action and the trivy binary release v0.69.4 were similarly backdoored. This was the second compromise in three weeks, following the hackerbot-claw incident on February 28. StepSecurity Harden-Runner detected the attack in real-time across 12,000+ monitored public repositories, and analysis confirmed 45 repositories executed the compromised action, with 5 having custom secrets including AWS credentials and Docker Hub tokens directly accessible. Homebrew executed an emergency downgrade to v0.69.3. Affected teams should rotate GITHUB_TOKEN and all explicitly referenced secrets immediately.
Updates to GitHub Copilot Interaction Data Usage Policy
Starting April 24, GitHub will use interaction data from Copilot Free, Pro, and Pro+ users to train and improve its AI models, unless users opt out via the Privacy settings page. The policy covers inputs, outputs, accepted or modified code suggestions, cursor context, file names, repository structure, and feedback ratings, and data may be shared with GitHub affiliates including Microsoft. Copilot Business and Enterprise users are not affected. GitHub says the decision is driven by measurable quality gains already observed from training on Microsoft employee interaction data, including higher suggestion acceptance rates across multiple programming languages. Users who previously opted out retain that preference automatically and will not be enrolled unless they actively opt in.
A Year of Open Source Vulnerability Trends: CVEs, Advisories, and Malware
GitHub published 4,101 reviewed advisories in 2025, the fewest since 2021, but this reflects a backlog depletion rather than improved security — newly reported vulnerability reviews actually rose 19% year over year. npm malware advisories surged 69% compared to 2024, driven largely by campaigns such as SHA1-Hulud, marking the highest count since malware tracking began in 2022. CWE-79 (cross-site scripting) remained the top weakness, while resource exhaustion (CWE-400, CWE-770), unsafe deserialization (CWE-502), and server-side request forgery (CWE-918) were unusually prevalent. Advisories missing any CWE classification dropped 85%, from 452 in 2024 to 65 in 2025. GitHub's CNA published 35% more CVE records than in 2024, outpacing the overall CVE Project's 21% growth, and 987 organizations submitted CVEs during the year.
Mozilla and Mila Announce Strategic Research Partnership to Advance Open Source and Sovereign AI
Mozilla and Mila, the Quebec Artificial Intelligence Institute, have announced a strategic research partnership — Mozilla's first-ever collaboration with a major AI research lab. The inaugural project focuses on trust and usability, including private memory architectures for AI agents, with the goal of demonstrating that open-source AI can match cutting-edge research capabilities. Mozilla President Mark Surman and Mila CEO Valerie Pisano both emphasized Canada's potential to lead on sovereign AI, citing the country's research depth and commitment to trustworthy, transparent systems. The partnership aims to reduce dependence on closed AI systems by developing technologies that enable transparency, accountability, and shared innovation across geographies, welcoming participation from research institutions and developers worldwide.
The Next.js Adapter API Just Shipped: What's Next for Multi-Platform Deployment
The Next.js Deployment Adapter API reached stable status in Next.js 16.2, a milestone achieved after 18 months of collaboration between Netlify, Cloudflare, Google/Firebase, AWS Amplify, OpenNext, and the Next.js team. Vercel's open-source adapter is already powering some production builds on their platform, while Netlify's and Cloudflare's adapters are in active development. The Next.js team has published the full 9,000+ end-to-end test suite as a conformance contract and established the Next.js Ecosystem Working Group with public meeting notes. Netlify's post also candidly details three architectural challenges the adapter API does not yet solve: Partial Prerendering (PPR) requires bespoke CDN infrastructure to stitch static and dynamic streams; cache revalidation demands atomic updates of coupled cache entries with no standard HTTP mechanism available; and on-demand revalidation is fundamentally incompatible with immutable CDN deploy models used by Netlify, Cloudflare Pages, and others.
Supply-chain security dominated week 13 with two critical incidents. First, the Trivy v0.69.4 release and all aquasecurity/trivy-action tags (0.0.1–0.34.2) were backdoored for 12 hours on March 19, harvesting CI/CD secrets via /proc/pid/mem and exfiltrating them to the typosquatted domain scan.aquasecurtiy.org. StepSecurity Harden-Runner detected the attack across 12,000+ monitored repositories, with 45 confirmed executions and 5 exposing AWS credentials and Docker Hub tokens. Homebrew emergency-downgraded to v0.69.3. The incident then escalated: researchers identified CanisterWorm, a self-propagating npm worm by the same TeamPCP threat actor, using stolen npm tokens to autonomously publish backdoored patch versions across 16+ packages in the @opengov scope, with a C2 endpoint on the ICP blockchain and a second-stage Kubernetes DaemonSet named host-provisioner-iran targeting every cluster node.
GitHub's annual open-source vulnerability report added context: npm malware advisories surged 69% in 2025 (driven by campaigns like SHA1-Hulud), newly reported vulnerabilities rose 19% year over year, and CWE-79 (XSS) remained the top weakness alongside resource exhaustion and unsafe deserialization. GitHub's CNA published 35% more CVE records than in 2024, outpacing the overall CVE Project's 21% growth.
Beyond security, GitHub announced that starting April 24 it will use interaction data from Copilot Free, Pro, and Pro+ users to train its models (opt-out available in Privacy settings; Business and Enterprise users are excluded). Mozilla and Mila announced a first-ever research partnership focused on open-source sovereign AI, starting with private memory architectures for AI agents. The AI frontier featured analysis of OpenAI's upcoming Spud model and a new ARC-AGI-3 benchmark where humans score 100% but frontier models including Gemini 3.1 score below 0.5%.
Key Takeaways
The Trivy supply-chain compromise spawned CanisterWorm — a self-propagating npm worm using a blockchain C2 and Kubernetes DaemonSet payload. Rotate all npm tokens and CI/CD secrets on any machine that ran affected packages, and check for ~/.local/share/pgmon/service.py.
npm malware advisories surged 69% in 2025 per GitHub's annual report; newly reported vulnerabilities rose 19% YoY. Supply-chain attacks are accelerating, not stabilizing.
GitHub Copilot Free/Pro/Pro+ interaction data will train AI models from April 24 — opt out via Privacy settings if needed; Business and Enterprise accounts are unaffected.
