CVE-2026-24887 · high
Claude Code confirmation-prompt bypass
Claude Code releases before 2.0.72.
Update to 2.0.72 or later and keep untrusted repository content outside command-authorizing context.
CapitalGuard Observatory · Tool record
Claude Code
Anthropic · coding agent
Claude Code is a local or cloud coding agent with file, command, network, MCP, and editing capabilities governed by permissions, sandboxing, trust, and account settings.
Public-Control Score
85
Extensive public control coverage
Measures publicly documented control coverage as of the review date. It is not a breach rate, penetration test, endorsement, or guarantee of safety.
Scored dimensions
The documented control surface.
This record is reviewed against the same five-dimension rubric used across every scored product.
Read MethodologyAccess and scope transparency
18/20Official material provides extensive operational detail for access and scope transparency.
Data and retention controls
16/20Official material names concrete controls and meaningful boundaries for data and retention controls.
Connector and permission controls
17/20Official material provides extensive operational detail for connector and permission controls.
Action, approval, and isolation controls
18/20Official material provides extensive operational detail for action, approval, and isolation controls.
Audit and administrative evidence
16/20Official material names concrete controls and meaningful boundaries for audit and administrative evidence.
Documented access points
1
repository and local file reads
2
edits and Bash commands
3
network access, MCP servers, hooks, and cloud environments
Settings to verify now
1
Permission mode and deny rules
2
Filesystem and network sandbox
3
Trusted directories, MCP servers, hooks, and unsandboxed escape paths
Linked security signals
Version checks that should not wait.
CGO-CHANGE-2026-06-09 · informational
GitHub adds security validation for third-party coding agents
Supported GitHub agent workflows; coverage depends on repository and agent settings.
Keep the validation enabled, but retain branch protection, review, and least-privilege controls around agent-generated changes.
CGO-RESEARCH-2026-05 · informational
AgentSecBench formalizes three enforceable agent-security boundaries
Agentic systems that combine untrusted content, private context, and action-capable tools.
Project trusted provenance, restrict capabilities before generation, and validate outputs before execution.
Linked evidence
Ten practical Claude Code checks.
Private file access
Claude Code Private File Access: What to Check
Credential exposure
Claude Code API Key Exposure: Check Before You Share
Client confidentiality
Claude Code Client Data Safety for Freelancers
Prompt injection
Claude Code Prompt Injection: A Practical Defense
Connector permissions
Claude Code Connector Permissions: Review the Real Scope
Command execution
Claude Code Command Execution: Keep Control
Unsafe generated code
Claude Code Generated Code Risks: Review Before You Run
History and sharing
Claude Code Chat History and Shared Links: Privacy Check
Accidental oversharing
Claude Code Sensitive Data: What Not to Paste
Autonomous actions
Claude Code Autonomous Actions: Approval Safety Guide