Backslash Identifies 30+ Security-Relevant Claude Code Fixes

Summary

Backslash Security says it found more than 30 security-relevant Claude Code fixes from April to early June 2026, including prompt injection and credential leaks. The report warns delayed updates may leave brief exposure windows.

Key points

Backslash Security reviewed Claude Code release logs and identified more than 30 security-relevant fixes shipped between April and early June 2026; Anthropic had not publicized them.
The issues included data poisoning, prompt injection, arbitrary code execution and OAuth credential leakage.
One flaw reportedly bypassed safeguards against destructive commands by adding a backslash; another allowed an AI agent to place a backdoor in shell startup files.
Backslash said Anthropic had addressed every issue identified in its report.
The report says frequent releases and concerns about stability or performance can lead organizations to delay updates, creating short periods of exposure.
Regulated or air-gapped environments, internal review processes and manual installations can make updates harder to apply promptly.

Article Details

Event Type: Backslash Security reported that Anthropic had patched more than 30 security-relevant issues in Claude Code between April and early June 2026, including prompt injection, data poisoning, arbitrary code execution, OAuth credential leakage, and a safeguard bypass for destructive commands.
Impact: The reported issues could expose data and credentials, enable malicious code or backdoors, or allow destructive commands. Delayed updates may leave organizations exposed while they balance security against stability, performance, and operational requirements. The article says Anthropic addressed every issue identified in the report.

People

Yossi PikCo-founder and chief technology officer of Backslash Security; discussed the report’s findings and the security implications of the rapid release cadence of AI tools.

Vendors

AnthropicBackslash Security found that Anthropic quietly patched more than 30 security-relevant issues in Claude Code between April and early June 2026, including data poisoning, prompt injection, arbitrary code execution, OAuth OpenAIThe report says the rapid release cadence of AI tools like Claude Code and OpenAI’s Codex creates short security gaps that force organizations to balance stability, performance, and protection.

Products

Claude CodeBackslash Security found that Anthropic quietly patched more than 30 security-relevant issues in Claude Code between April and early June 2026, including data poisoning, prompt injection, arbitrary code execution, OAuth CodexThe report says the rapid release cadence of AI tools like Claude Code and OpenAI’s Codex creates short security gaps that force organizations to balance stability, performance, and protection.