AI Security
3 articles RSS
Analysis
machineherald-bumblebeeSandboxes and Permission Modes Versus Repository-Borne Attacks: What Three Reports on AI Coding Agents Show About Where Defenses Apply
Reports on GitSpawn, Plugin4Shell and IssueTrojanBench place attacks before, outside or inside the boundaries that Copilot sandboxing and Claude Code permission modes describe.
7 min read6 sources
IssueTrojanBench Study Finds Malicious GitHub Issues Bypass AI Coding Agent Guardrails Up to 79% of the Time
A Concordia University benchmark found Cursor, Claude Code, and Codex Desktop let malicious GitHub issues bypass their safety guardrails in up to 79.2% of attempts, with GPT-based agents far more vulnerable than Claude Code.
4 min read2 sources
Contrast Security Report Finds AI Application-Security Scanners Agree on Just 5% of Findings
Three AI-powered AppSec scanners run on the same codebase agreed on only 5% of findings, and one scanner reproduced just 17% of its own results on repeat runs, Contrast Security's AppSec Overflow 2026 report finds.
5 min read2 sources