--- title: "#ai-security articles | Hashnode" description: "Explore 437 articles tagged #ai-security on Hashnode. Tutorials, guides, and how-tos from the community." url: "https://hashnode.com/tag/ai-security" --- Tag # #ai-security 437 posts · 2 followers [404 Founders](https://hashnode.com/@404-founders) · [6h ago](https://hashnode.com/posts/claude-code-settings-local-json-secrets-leak/6af471b8b7598182981c90fc) ### [Claude Code settings.local.json: How Approved Commands Leak Secrets](https://404-founders.com/blog/claude-code-settings-local-json-secrets-leak?utm_source=hashnode&utm_medium=feed) The approval rule stores the whole command, and the file that keeps it out of git lives on one laptop 12 min read · [404-founders.com](https://404-founders.com/blog/claude-code-settings-local-json-secrets-leak?utm_source=hashnode&utm_medium=feed) [Damiano Bianda](https://hashnode.com/@damianobianda) · [8h ago](https://hashnode.com/posts/seven-neurons-and-a-trigger-string-what-microsoft-s-lazarus-backdoor-actually-proves/6af43c78d48bcb5188563a40) ### [Seven Neurons and a Trigger String: What Microsoft's Lazarus Backdoor Actually Proves](https://damianobianda.hashnode.dev/seven-neurons-and-a-trigger-string-what-microsoft-s-lazarus-backdoor-actually-proves?utm_source=hashnode&utm_medium=feed) The headline writes itself: Microsoft researchers backdoored an AI model by editing fewer than 10 neurons, no retraining required. It sounds like the opening of a disaster movie. The research, publish 6 min read · [damianobianda.hashnode.dev](https://damianobianda.hashnode.dev/seven-neurons-and-a-trigger-string-what-microsoft-s-lazarus-backdoor-actually-proves?utm_source=hashnode&utm_medium=feed) [404 Founders](https://hashnode.com/@404-founders) · [13h ago](https://hashnode.com/posts/bossai-ai-attack-mesh-mcp-tool-servers/6af41ea5e3a1a177c41abcb9) ### [BOSSAI AI Attack Mesh: One Operator, 12 Agents and Open Tool Servers](https://404-founders.com/blog/bossai-ai-attack-mesh-mcp-tool-servers?utm_source=hashnode&utm_medium=feed) One operator ran 12 AI agents against Taiwan's emergency services. How the BOSSAI mesh worked, why its MCP tool servers matter, and how to detect it. 17 min read · [404-founders.com](https://404-founders.com/blog/bossai-ai-attack-mesh-mcp-tool-servers?utm_source=hashnode&utm_medium=feed) [Fonz](https://hashnode.com/@aiza-hextyx) · [3d ago](https://hashnode.com/posts/an-openai-model-bypassed-its-sandbox-s-network-restrictions-here-s-why-it-matters/6af06dc36bd10e54d9921901) ### [An OpenAI Model Bypassed Its Sandbox's Network Restrictions. Here's Why It Matters.](https://aiza-hextyx.hashnode.dev/an-openai-model-bypassed-its-sandbox-s-network-restrictions-here-s-why-it-matters?utm_source=hashnode&utm_medium=feed) A research model couldn't use its approved internet connection, so it wrote its own networking code. The incident exposes a key risk in AI security. 11 min read · [aiza-hextyx.hashnode.dev](https://aiza-hextyx.hashnode.dev/an-openai-model-bypassed-its-sandbox-s-network-restrictions-here-s-why-it-matters?utm_source=hashnode&utm_medium=feed) [404 Founders](https://hashnode.com/@404-founders) · [3d ago](https://hashnode.com/posts/langflow-url-component-vulnerability-ghsa-qvqp-mxh4-f27w/6af0810bd008cd2df64916f9) ### [Langflow URL Component Vulnerability: Old Flows Keep Old Code](https://404-founders.com/blog/langflow-url-component-vulnerability-ghsa-qvqp-mxh4-f27w?utm_source=hashnode&utm_medium=feed) GHSA-qvqp-mxh4-f27w: one chat message can exhaust Langflow memory via the URL agent tool. Why 1.13.0 may not fix saved flows, and how to check. 16 min read · [404-founders.com](https://404-founders.com/blog/langflow-url-component-vulnerability-ghsa-qvqp-mxh4-f27w?utm_source=hashnode&utm_medium=feed) [Fonz](https://hashnode.com/@aiza-hextyx) · [4d ago](https://hashnode.com/posts/the-6-3-billion-binance-hack-why-devops-workstations-shouldn-t-have-unlimited-internet-access/6aef29c22c305a747a6631a6) ### [The $6.3 Billion Binance Hack: Why DevOps Workstations Shouldn't Have Unlimited Internet Access](https://aiza-hextyx.hashnode.dev/the-6-3-billion-binance-hack-why-devops-workstations-shouldn-t-have-unlimited-internet-access?utm_source=hashnode&utm_medium=feed) The $6.3B Binance hack exposes unrestricted DevOps internet access risks. Network isolation, egress controls, and secure build environ reduce risk. 12 min read · [aiza-hextyx.hashnode.dev](https://aiza-hextyx.hashnode.dev/the-6-3-billion-binance-hack-why-devops-workstations-shouldn-t-have-unlimited-internet-access?utm_source=hashnode&utm_medium=feed) [Ivan Klawd](https://hashnode.com/@ivanklawd) · [5d ago](https://hashnode.com/posts/gemini-4-carbon-gated-frontier-access/6aeceeb5849c49b0637fdf08) ### [Gemini 4 Carbon: Gated Frontier Access Is the New Norm](https://blog.thecgaigroup.com/gemini-4-carbon-gated-frontier-access?utm_source=hashnode&utm_medium=feed) Why staged releases, pre-release testing and price step-ups should change how enterprises plan model adoption 14 min read · [blog.thecgaigroup.com](https://blog.thecgaigroup.com/gemini-4-carbon-gated-frontier-access?utm_source=hashnode&utm_medium=feed) [404 Founders](https://hashnode.com/@404-founders) · [5d ago](https://hashnode.com/posts/prompt-adware-codex-jailbreak-llm-gateway/6aee07e82c6bb12cf9ee8c32) ### [Prompt Adware: A Codex Jailbreak Bundle That Sells an LLM Gateway](https://404-founders.com/blog/prompt-adware-codex-jailbreak-llm-gateway?utm_source=hashnode&utm_medium=feed) Haiou rewrites OpenAI Codex global instructions to advertise an LLM gateway. How prompt adware works, why the gateway is the risk, and what to check. 16 min read · [404-founders.com](https://404-founders.com/blog/prompt-adware-codex-jailbreak-llm-gateway?utm_source=hashnode&utm_medium=feed) [404 Founders](https://hashnode.com/@404-founders) · [6d ago](https://hashnode.com/posts/mcp-server-path-traversal-aws-serverless-cve-2026-116635/6aec13f337f714f71ab4543a) ### [MCP Server Path Traversal: CVE-2026-116635 in AWS Serverless MCP](https://404-founders.com/blog/mcp-server-path-traversal-aws-serverless-cve-2026-116635?utm_source=hashnode&utm_medium=feed) CVE-2026-116635 lets a crafted project name escape the AWS Serverless MCP Server's status folder. Why the write half matters and what to check. 17 min read · [404-founders.com](https://404-founders.com/blog/mcp-server-path-traversal-aws-serverless-cve-2026-116635?utm_source=hashnode&utm_medium=feed) [404 Founders](https://hashnode.com/@404-founders) · [6d ago](https://hashnode.com/posts/vllm-cors-vulnerability-cve-2026-115917/6aec14525705da2d5d900a6c) ### [vLLM CORS Vulnerability: The API Key Never Covered /invocations](https://404-founders.com/blog/vllm-cors-vulnerability-cve-2026-115917?utm_source=hashnode&utm_medium=feed) vLLM before 0.32.0 let any web page read model output via the browser, even with --api-key set. Checks and fixes for CVE-2026-115917. 15 min read · [404-founders.com](https://404-founders.com/blog/vllm-cors-vulnerability-cve-2026-115917?utm_source=hashnode&utm_medium=feed)