Unaddressed vulnerability in Claude models
Security researcher Malinor disclosed a Cross-Model Context Inheritance vulnerability in Claude models. The bug allows bypassing safety filters to generate prohibited content, including harmful material.
Malinor reported the vulnerability to Anthropic on February 17, 2026, using 14 communication channels, including US regulators. The company sent two canned responses, dismissing it as a standard feedback report, and did not fix the issue.
In April, Anthropic released Claude Opus 4.7, which still exhibits the vulnerability. The researcher’s account wasn’t even suspended for generating prohibited content. The public disclosure highlights serious shortcomings in Anthropic’s safety processes and feedback loop.
Up next
Claude Code v2.1.147 released
Claude Code adds Workflow, a tool for orchestrating multi-agent systems. This lets you control and coordinate specialized sub-agents. The feature is opt-in for now; set CLAUDE_CODE_WORKFLOWS=1 to try it.
Related stories
- Apiiro finds Claude Code secrets in 1,122 GitHub files
- Email leak stopgap in Claude Code checks only shell calls
- Claude Code agent sat on a Pentagon network for two months
- Irregular CEO blames human error for AI agent breaches
- FlashProve proofs cost ~70-95x the GPU time of inference
- Backdoored Claude Code builds were limited to Cursor
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.