Sebastion
←All topics

ai

41 pieces of writing

OpenAI's Hugging Face incident was not an AI escape story. It was a sandbox failure.
security12 min read

OpenAI's Hugging Face incident was not an AI escape story. It was a sandbox failure.

OpenAI and Hugging Face disclosed an AI cyber-evaluation incident with real security impact. The useful lesson is not that AI escaped, but that eval sandboxes need hard containment.

security13 min read

CVE-2026-21440 and AI file tools show path traversal is an architectural antipattern

MarSeventh/CloudFlare-ImgBed PR #625 blocked SSRF in fetchRes
vulnerability8 min read

MarSeventh/CloudFlare-ImgBed PR #625 blocked SSRF in fetchRes

security12 min read

AI agents turned poisoned repositories and obfuscated code into supply-chain execution paths

AI agents changed repository poisoning from a code-review problem into an execution problem because models now read untrusted project text, interpret obfuscated code and act through trusted developer tooling.

security13 min read

Amazon Q, Claude Code and MCP made repository text a supply-chain attack surface

security13 min read

OpenClaw's 470 advisories show unauthenticated RCE became a cloud AI platform pattern

security13 min read

CIFSwitch CVE-2026-46243 and PraisonAI show privilege escalation is an architectural antipattern

CIFSwitch CVE-2026-46243 and PraisonAI show why vertical movement often follows from designs that let low-trust identities shape high-trust operations.

security12 min read

Gogs, PraisonAI and KnowledgeDeliver show authentication bypass is a self-hosted platform design failure

MCP OAuth token persistence turns AI orchestration into a supply-chain trust boundary
security13 min read

MCP OAuth token persistence turns AI orchestration into a supply-chain trust boundary

GitHub Actions OIDC and TanStack show why 2026 supply chain attacks target release authority
security11 min read

GitHub Actions OIDC and TanStack show why 2026 supply chain attacks target release authority

Supply chain compromise has shifted from stealing credentials to poisoning package ecosystems through compromised CI/CD systems, maintainer accounts and trusted execution paths.

maboloshi/github-chinese PR #692 fixed DOM XSS in translation results
vulnerability8 min read

maboloshi/github-chinese PR #692 fixed DOM XSS in translation results

RAGFlow PR #14803 removed a CWE-502 pickle RCE footgun from deserialize_b64
vulnerability7 min read

RAGFlow PR #14803 removed a CWE-502 pickle RCE footgun from deserialize_b64

getsentry/XcodeBuildMCP PR #289 hardens shell escaping in MCP tool execution
vulnerability9 min read

getsentry/XcodeBuildMCP PR #289 hardens shell escaping in MCP tool execution

getsentry/XcodeBuildMCP accepted MCP tool parameters that could reach /bin/sh -c through unsafe double-quote escaping. PR #289 replaces that path with POSIX single-quote escaping and adds regression coverage.

AReaL PR #1323 refuses the default admin API key on network-facing proxy binds
vulnerability9 min read

AReaL PR #1323 refuses the default admin API key on network-facing proxy binds

Harbor PR #236 blocks CWE-78 in remote profile downloads
vulnerability7 min read

Harbor PR #236 blocks CWE-78 in remote profile downloads