Prompt injection remains an unfixed structural weakness in AI agents because large language models cannot distinguish developer instructions from text they read, and indirect prompt injection — hidden…
Security researcher Johann Rehberger reported a 60-80% code execution success rate against Claude Code Opus 5's Auto Mode, bypassing a Trajectory Labs eval that had rated it at 0.00% prompt injection …
Security researcher Johann Rehberger published a multi-stage attack on August 26 achieving remote code execution against Claude Code Opus 5 in Auto Mode with a 60 to 80 percent success rate, exploitin…
Anthropic's Claude Code Auto Mode, marketed as a protective layer against prompt injection attacks, can be bypassed and even blocks the agent's own cleanup commands, according to research by Johann Re…
Anthropic's Claude Code running Opus 5 in Auto Mode can be tricked into executing attacker-controlled code up to 80 percent of the time by asking it to summarize a malicious website, according to rese…
Security researcher Johann Rehberger, writing on his blog Embrace The Red, published an exploit that tricks Anthropic's Claude Code into downloading and executing malware while running in Auto Mode, s…
Security researcher Johann Rehberger disclosed a chain of vulnerabilities in Microsoft Copilot, assigned CVE-2026-24299, that allowed persistent data exfiltration via HTML preview, delayed tool invoca…
Independent security researcher Ayush Paul demonstrated a link-walking exploit against Anthropic's Claude that exfiltrates user memory data by encoding it into URL paths, while a second researcher, Jo…
Renuo developers identified security risks in AI coding agents like Claude Code and OpenCode, including credential exfiltration and destructive actions via prompt injection, after incidents such as un…