AppSec Weekly Content Briefing
October 29, 2025

AI Writes Code Faster Than Security Teams Can Fix It

6 min

TL;DR

AI coding tools are skipping software best practices, and new "prompt injection" attacks are exploiting agent workflows in ways your current scanners can't even see. Ox Security analysis
Five critical CVEs exploited in the wild while organizations race to patch. Remediation windows are now measured in days, not weeks. Windows WSUS | Adobe Commerce
Path traversal bug compromised 3,000+ MCP servers, exposing thousands of API keys in the largest AI supply chain attack yet. MCP server breach
This week: 8 vulnerabilities disclosed | 5 actively exploited | 3,000+ servers compromised

Big Picture

The math doesn't work anymore. Every efficiency gain in development, whether from AI code generation, automated deployment, or continuous delivery, creates security debt that compounds faster than humans can triage. Meanwhile, attackers turned the patch window into a race they're winning. This week shows the collision.

AI Code Generation Outpaces Security Remediation Capacity

The timing this week couldn't be more stark. Just as GitHub launched Agent HQ, new research landed showing just how risky AI-generated code can be. An Ox Security analysis of 300+ repositories revealed that AI tools systematically violate at least 10 software engineering best practices. Worse, Trail of Bits demonstrated how AI agents running shell commands can be tricked into executing malicious code—an attack surface your SAST/DAST tools will miss entirely.

The CrowdStrike survey quantifies the pain: 76% of organizations are struggling. AI is creating a massive bottleneck right where you live—in the remediation pipeline. With threats like the GlassWorm malware now actively spreading through VS Code extensions, it's clear attackers are already targeting this new gap.

Takeaways

You're facing a capacity crisis you can't hire your way out of. AI tools are generating code that ignores engineering best practices, creating new attack surfaces faster than your manual triage-and-remediate workflows can possibly handle.

Critical Vulnerabilities Exploited Faster Than Patches Can Deploy

It was a brutal week for patch-and-pray. Five critical vulnerabilities were actively exploited while patch deployment lagged behind.

The Windows WSUS vulnerability (CVE-2025-59287) was weaponized almost immediately after disclosure. The Adobe Commerce "SessionReaper" bug is actively targeting customer accounts on vulnerable sites. CISA's emergency Nov. 12 deadline for the Lanscope CVE shows how fast response windows are closing.

The pipeline is getting fuller. Pwn2Own just dropped 56 new zero-days into the 90-day vendor pipeline, creating a predictable wave of patches coming your way. The gap between disclosure and exploitation continues to shrink.

But this constant firefighting is distracting from a deeper, more systemic risk: the supply chain itself is under attack.

Takeaways

Attackers are weaponizing CVEs in hours, not weeks. Enterprise patch cycles that worked last year are struggling to keep pace. With regulatory deadlines shrinking to days, MTTR is becoming a more critical metric for teams managing this volume.

Supply Chain Attacks Escalate Against AI Development Infrastructure

The big story this week is the path traversal vulnerability in Smithery.ai. The attack compromised over 3,000 MCP servers and exposed thousands of API keys—one of the largest AI supply chain attacks on record. MCP is the framework GitHub just announced for its new AI agent tooling.

This isn't an isolated incident. The Lazarus Group is targeting defense contractors through compromised open-source packages. GlassWorm's self-propagating worm is spreading through developer AI tools. The attack surface is expanding beyond traditional software dependencies into the AI development stack itself.

As these supply chain threats escalate, the broader security market is consolidating around AI-powered approaches.

Takeaways

Supply chain attacks are targeting AI development infrastructure. Teams adopting MCP workflows or AI coding tools are inheriting new exposure in their dependency chains. Third-party dependencies are becoming a more active threat vector, not just a compliance concern.

The AI Security Market Consolidates Around Orchestration

The market moved fast this week. Snyk launched Evo as the "world's first Agentic Security Orchestrator." Opsera's Hummingbird AI Agent appeared in GitHub's official MCP registry. Red Hat Developer Lightspeed expanded its AI workflows. Four major vendors launched AI-focused security platforms in a single week.

The CrowdStrike survey data aligns with this market activity—76% of teams can't keep pace with current remediation workflows.

Takeaways

The vendor landscape is consolidating around AI-powered approaches.

Vulnerabilities in the Wild

CVE-2025-59287Product: Windows Server WSUS — CVSS: Critical — Impact: Remote Code Execution — Status: Actively Exploited

CVE-2025-54236Product: Adobe Commerce / Magento (SessionReaper) — CVSS: Critical — Impact: Remote Code Execution — Status: Actively Exploited

CVE-2025-61932Product: Lanscope Endpoint Manager — CVSS: Critical — Impact: Zero-day exploitation — Status: Actively Exploited, added to CISA KEV (deadline: Nov 12)

CVE-2025-33073Product: Windows NTLM — CVSS: High — Impact: Authentication bypass — Status: PoC Available

TARmageddonProduct: Rust async-tar library — CVSS: Critical — Impact: Remote Code Execution in abandoned dependency — Status: PoC Available

CVE-2025-55315Product: Kestrel ASP.NET Core — CVSS: 9.9 (Critical) — Impact: HTTP request smuggling enables credential hijacking — Status: Patch Available

MCP Server Path TraversalProduct: Smithery.ai (3,000+ MCP servers) — CVSS: Critical — Impact: API key exposure affecting thousands of servers — Status: Fixed (June 2025), disclosed Oct 2025

GlassWormProduct: VS Code Extensions — CVSS: High — Impact: Self-propagating worm, first supply chain attack on developer AI tools — Status: Actively Exploited

Your Curated Weekly Reading List

This Week's Headlines

Critical Windows Server WSUS Vulnerability Exploited in the WildWhy it's worth your time: Windows WSUS CVE-2025-59287 exploited quickly after patch release shows how attackers weaponize disclosures faster than enterprises deploy fixes.

Exploitation of Critical Adobe Commerce Flaw Puts Many eCommerce Sites at RiskWhy it's worth your time: SessionReaper actively exploited against Magento stores while many remain unpatched despite targeting customer accounts.

Lanscope Endpoint Manager Zero-Day Exploited in the WildWhy it's worth your time: CISA added to KEV with November 12 federal deadline, accelerating remediation timelines.

TARmageddon Flaw in Popular Rust Library Leads to RCEWhy it's worth your time: Abandoned async-tar library highlights supply chain challenge of unmaintained dependencies.

Self-propagating worm found in VS Code extension marketplacesWhy it's worth your time: GlassWorm is first major supply chain attack targeting developer AI tooling.

Pwn2Own Day 2: Hackers exploit 56 zero-days for $790,000Why it's worth your time: 56 zero-days with 90-day timelines create predictable patching waves requiring elastic capacity.

Deep Dives & Analysis

OpenAI Atlas Omnibox Is Vulnerable to JailbreaksWhy it's worth your time: Trail of Bits showed spoofed AI sidebars trick Atlas users into dangerous actions.

Analysis Identifies 10 AI Coding Tool Behaviors That Ignore Best Software Engineering PracticesWhy it's worth your time: Ox Security analyzed 300+ repositories showing AI tools create systematic technical debt.

DevSecCon 2025 Recap: Securing the AI Revolution TogetherWhy it's worth your time: Snyk launched Evo as 'Agentic Security Orchestrator,' shifting competitive positioning.

Survey: Cybersecurity Teams Struggling to Keep Pace in the Age of AIWhy it's worth your time: 76% struggling quantifies capacity crisis automated remediation solves.

From Path Traversal to Supply Chain Compromise: Breaking MCP Server HostingWhy it's worth your time: 3,000+ MCP servers compromised affects enterprises adopting GitHub MCP ecosystem.

Red Hat Developer Lightspeed brings AI assistance to Developer HubWhy it's worth your time: Traditional vendors entering AI-native development space.

Opsera Adds AI Agent for Code Quality AnalysisWhy it's worth your time: Hummingbird in GitHub's MCP registry indicates ecosystem consolidation.

Prompt injection to RCE in AI agentsWhy it's worth your time: Trail of Bits demonstrates new attack surface category for AI agent command execution.

Lazarus group targets European drone makersWhy it's worth your time: State-sponsored actors weaponize supply chain targeting defense contractors.

UN agreement on cybercrime criticized over researcher risksWhy it's worth your time: Treaty discussions create compliance complexity alongside faster patching requirements.


Looking to Stay Up to Date with All Things AppSec?

Subscribe to the Weekly AppSec Briefing and never miss a thing.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.