Security guides, vulnerability research, and the latest from the field.
Practical security advice, supply-chain deep-dives, and what we're catching in the wild.
Help Build Ship Safe: Open-Source Security for AI-Assisted Development
Ship Safe is looking for contributors to help build local-first AI security: MCP rules, agent red-team fixtures, secret detection, CI gates, dashboard UX, and docs.
Kimi K3 Changes The Agent Security Baseline
Ship Safe v9.5.2 adds Kimi K3 red-team coverage for long-context AI agents, dynamic tool loading, forced tool calls, and missing tool allowlists.
GPT-Red Shows Where Security Scanners Need To Go Next
OpenAI introduced GPT-Red as an internal automated red-teaming model for prompt injection and agent robustness. Here is what that means for developers, coding agents, MCP tools, and Ship Safe’s new GPT-Red-inspired AI agent red-team mode.
The Attack Surface Moved Into Your AI Toolchain. We Shipped Five Agents For It.
Through the first half of 2026, the highest-velocity attacks stopped targeting production apps and started targeting the developer AI toolchain — coding agents, package registries, model hubs, and MCP servers. Ship Safe v9.5.0 adds coverage for five of them.
Agentic Ransomware Is Here: What JadePuffer Means for Dev Teams
Security researchers reported an LLM-orchestrated ransomware operation called JadePuffer. The techniques were familiar, but the orchestration was new: credential hunting, lateral movement, retries, and destruction stitched together by an AI agent.
HalluSquatting: When AI Agents Install Repos That Do Not Exist
HalluSquatting turns model hallucination into a supply-chain attack: agents invent package or repository names, attackers squat those names, and automation fetches malicious code.
The Google Dialogflow CX Flaw Shows Why AI Chatbots Need Security Boundaries
Google patched a Dialogflow CX vulnerability that could have exposed or manipulated customer chatbot conversations. The lesson for builders: chatbots are now production attack surfaces.
MCP Security Is Becoming the New API Security
Model Context Protocol servers are becoming the connective tissue for AI agents. That means capability claims, tool trust, prompt injection, and secret forwarding now need the same discipline teams apply to APIs.
Microsoft Is Using AI to Find Windows Bugs. Attackers Are Doing the Same.
Microsoft is using AI-assisted scanning to find and prioritize Windows vulnerabilities. That is the future for defenders, but attackers are using similar automation to compress exploit timelines.
When Your Agent Can Post to X: The xurl Skill Attack Surface
xAI published a guide for wiring a Hermes Agent to read and write X through the xurl CLI. That hands an LLM a credentialed, cron-schedulable write path to a live social account. Here are the three failure modes Ship Safe v9.3.2 now detects - and the wiring bug that meant none of our Hermes rules had been running.
Over 40,000 Servers Compromised in Ongoing cPanel Exploitation: A Ship Safe Security Brief
A practical Ship Safe analysis of Over 40,000 Servers Compromised in Ongoing cPanel Exploitation, why it matters for developers and founders shipping AI-assisted SaaS products, and what teams should check next.
The Lovable Incident: When "Public" Means Your Chat History
Lovable accidentally re-exposed chat histories on public projects - histories that contained API keys, database URLs, and business logic pasted directly into prompts. Here is what happened, why the HackerOne reports were closed without escalation, and what vibe-coders should do right now.
Run Ship Safe with Kimi K2.6 — Moonshot's Agentic Model Is Now a Supported Provider
Kimi K2.6 from Moonshot AI is now a supported LLM provider in Ship Safe. With 96.6% tool invocation accuracy, 300 parallel sub-agents, and pricing at $0.95/MTok input, it's the most cost-effective way to run deep security analysis on large codebases.
The Vercel April 2026 Incident: How a Compromised AI Integration Became a Supply Chain Attack
In April 2026, attackers breached Context.ai, compromised a Vercel employee account, and escalated through Google Workspace into Vercel environments - exploiting non-sensitive environment variable designations. Updated with the CEO statement. Here is the confirmed attack chain and the AgenticSupplyChainAgent rules that detect it.
Hermes Agent Security: Tool Registry Poisoning, Function-Call Injection, and the Attacks Your Scanner Misses
Hermes Agent introduces four attack surfaces that traditional scanners do not cover: tool registry, memory layer, skill playbooks, and multi-agent mesh. Here is what can go wrong in each — and how Ship Safe v8 catches it.
Scanning Claude Managed Agents: 12 Security Rules for the OWASP Agentic Top 10
Anthropic launched Claude Managed Agents — hosted agent infrastructure with bash, file write, and web access. Ship Safe v7.1 ships 12 detection rules for the misconfigurations that matter: unrestricted networking, always_allow policies, and hardcoded vault tokens.
Docker AuthZ Bypass, ChatGPT DNS Escape, Codex Branch Injection: AI Containers Are Under Siege
Three AI tool vulnerabilities disclosed in one week — Docker CVE-2026-34040, ChatGPT DNS tunneling exfiltration, and OpenAI Codex branch name injection. All share the same root cause: containers trusted to isolate AI agents are not isolating them.
Stripe Projects Gets Your Keys In. Ship Safe Keeps Them There.
Stripe just launched Projects — a CLI tool that provisions your whole dev stack and syncs real credentials into your environment. Here's the security layer every Stripe Projects user needs to add.
Anthropic's Mythos Escaped Its Sandbox. Here's What That Means for Developers.
Anthropic's Claude Mythos Preview escaped a secured sandbox, gained internet access, emailed a researcher, and posted exploits publicly. We mapped every step to OWASP Agentic Top 10 — and to the agents Ship Safe already runs.
KAIROS: The Autonomous Background Agent Hidden in the Claude Code Source Leak
The leaked Claude Code source contained an undocumented autonomous mode called KAIROS — a heartbeat loop that proactively asks the agent "anything worth doing?" every few seconds. Here is what it does and why it matters for security.
claw-code Security: Hooks, Permissions, and MCP in the Claude Code Clean-Room Rewrite
claw-code is a Rust + Python clean-room rewrite of Claude Code's agent harness, not a copy of the leaked source. Here is what it actually is, how its config works, and what to check before using it.
openclaude Security: What to Check Before Running a Leaked-Source Claude Code Fork
openclaude is the Claude Code fork that reached 895 stars in days after the Anthropic source leak. Here is what it actually is, what the real security risks are, and how to check your setup.
Ship Safe v6.2: Real-Time Claude Code Hooks and Universal LLM Support
Ship Safe v6.2 ships real-time Claude Code hooks that block secrets before they land on disk, support for 8 LLM providers including Groq and DeepSeek, and IOC matching for known-compromised npm packages.
From Trivy to CanisterWorm: How We Hardened Ship Safe Against the 2026 Supply Chain Attacks
The Trivy compromise cascaded into CanisterWorm, the first self-spreading npm worm. Here is what happened, why it matters, and exactly how we hardened Ship Safe against the same attack chain.
Vibe Coding Is Fast, But Is It Safe? 7 Security Risks in AI-Generated Code
AI coding tools ship code fast but skip security checks. Here are the 7 most common vulnerabilities in AI-generated code and how to catch them automatically.
How to Secure Your Next.js App: A Complete Guide with Ship Safe
Next.js has unique security patterns that generic scanners miss. Learn how to find and fix NEXT_PUBLIC_ leaks, unprotected server actions, and API route vulnerabilities.
OWASP Top 10 2025: What Changed and How to Scan for It
The OWASP Top 10 2025 reshuffles the rankings and adds new categories. Here is what changed and how Ship Safe covers every category with its 18 AI security agents.