Anthropic Launches Claude Security Plugin to Let Claude Review Its Own Code in Terminal Workflow
Anthropic has introduced the Claude Security plugin in beta, integrating it directly into the Claude Code terminal workflow. Developers who finish a feature with Claude Code now see an option to let the AI scan the code for vulnerabilities it may have introduced before committing.
How the Plugin Operates
Traditional rule-based scanners rely on matching known vulnerability patterns, which produces high false-positive rates and misses logic flaws. Claude Security instead uses a multi-agent system: it first reads the code and maps the architecture, then identifies potential threat points, and finally verifies suspicious findings to filter out false positives. The plugin generates patch suggestions that match the project’s existing coding style rather than generic alerts.
Every proposed fix must be reviewed and approved by a human; the tool never commits changes automatically. This design keeps the developer in control while adding an extra layer of analysis inside the existing command-line workflow.
Addressing Gaps in Traditional Tools
The plugin targets severe vulnerability classes that rule-based scanners often overlook, including memory corruption, injection defects, authentication bypasses, and complex logic errors. It excels at cross-file issues where data flows from one file to another and is incorrectly trusted in a business branch. By reading Git history and tracing data flows across files, the system reasons about whether a potential issue is actually exploitable instead of applying mechanical pattern matching.
Limitations and Practical Considerations
The feature remains in beta, and Anthropic has not published quantitative data on false-positive or false-negative rates. Large full-repository scans can consume significant tokens and occasionally switch to more expensive models, so teams are advised to limit scans to changed directories or use scheduled incremental checks. The company positions the tool as a supplement to existing SAST, DAST, and human security processes rather than a replacement.
Recommended Daily Usage
The most cost-effective approach is to run the scanner on uncommitted diffs before each commit. Results can be forwarded via webhook to Slack or Jira, exported as CSV or Markdown for audit trails, and dismissed findings are tracked to prevent repeated review. This workflow raises the security baseline for small teams and open-source maintainers who lack dedicated security staff.
Broader Implications of AI Reviewing AI-Generated Code
The release highlights an emerging pattern in which the same AI capabilities used to generate code are now applied to detect vulnerabilities created by that generation process. While the approach can surface issues traditional tools miss, Anthropic emphasizes that human oversight remains essential and that token costs must be managed. The long-term impact on code security will depend on real-world performance data gathered during the beta period.
Related articles
Agent-Ops 0.4.0 Released: Methodology for Secure Human-AI Collaboration in IT Operations
Sergey Zhitinsky, founder of Git in Sky, has published the public normative candidate for Agent-Ops 0.4.0, an open industry methodology governing how engineers and AI agents jointly handle IT infrastructure tasks. The framework keeps humans firmly in the decision-making loop while using deterministic programs for data collection and approved changes. It addresses risks such as prompt injection through processed data, unverified model outputs, and unclear accountability when AI recommendations lead to incidents. The methodology divides work across eight explicit steps and three separate planes: data, governance, and independent verification performed by a Guardian role. Two additional companies have joined as maintainers following agreements at the IT Elements 2026 conference, turning the project into a multi-organization effort. Contributors are invited to help refine contracts, schemas, and operational scenarios through GitHub and GitVerse.
ProxyKey MCP: Securing API Access for AI Agents Without Exposing Credentials
ProxyKey has released an MCP server that allows AI coding agents such as Claude Code and Cursor to manage API credentials without ever reading the actual secret values. The solution addresses the risk that any key visible to an agent becomes compromised through logging, tracing, or prompt injection. Real provider keys are stored encrypted with AES-256-GCM and never returned by any API endpoint after initial entry. Agents instead receive limited virtual passes that support IP binding, rate limits, TTL, and detailed request logging. A pending-secret workflow lets agents prepare services before the real token exists, with the human entering the secret only through a web panel. The approach deliberately restricts the MCP tool contract so no operation can read or return secret values.
Shadow AI in CI/CD: Why AI Agents Must Be Modeled as Security Threats
A new analysis from the CNCF highlights the growing risks of Shadow AI within continuous integration and continuous deployment pipelines. The report argues that AI agents should be treated as potential threats rather than simple productivity tools. Starting from a developer's laptop and extending to Kubernetes clusters, these agents can introduce unauthorized access paths and data exposure risks. Security teams are urged to incorporate AI agent behavior into formal threat modeling exercises. The discussion emphasizes the need for visibility and control over autonomous AI components operating in production environments.
Detecting Lateral Movement with Neural Networks Trained Solely on Synthetic Data
A researcher generated entire corporate network histories using a 135-line configuration file to create synthetic authentication logs containing lateral movement attacks. Neural networks trained exclusively on these artificial datasets were then evaluated against 1.65 billion real authentication events from Los Alamos National Laboratory, including 749 red team events across 301 compromised machines. The best ensemble of six models flagged 3.6 million hourly machine windows and placed 16 genuine attacks among the top 23 highest-scoring entries, producing only seven false positives. In comparison, a simple threshold counter required 161,000 false alarms to reach the same detection level. The approach also demonstrated an iterative feedback loop where detector errors directly informed refinements to the synthetic world generator. The work shows that synthetic data can reach AUC performance comparable to models trained on real labeled attacks while providing full control over the underlying attack definitions.