Benchmark: AI Agent Action Gating Response Times
15RL benchmarks AI agent action gating latency. SafeClaw adds 0.3ms median overhead. Full methodology and results for file, network, and shell gating.
Research: The Risk Curve of AI Agent Autonomy
15RL models the relationship between AI agent autonomy levels and operational risk, identifying critical thresholds where safety controls become essential.
15RL Study: The True Cost of Uncontrolled AI Agent Spending
15RL study finds uncontrolled AI agents cause median cost overruns of 280%. Analysis of 156 incidents with budget control recommendations.
15RL Incident Report: What Goes Wrong Without Agent Safety
15 Research Lab analyzes 214 AI agent safety incidents. Common failure modes include data exfiltration, resource exhaustion, and unauthorized file access.
15RL Framework: AI Agent Safety Maturity Model
15RL introduces a five-level maturity model for AI agent safety, helping organizations assess their current posture and plan systematic improvements.
A Taxonomy of AI Agent Risks: 15RL Classification Framework
15 Research Lab presents a structured taxonomy of AI agent risks across file, network, shell, and data categories with severity ratings and mitigations.
15RL Analysis: AI Agent Safety Tools Compared
15 Research Lab compares leading AI agent safety tools across action gating, latency, auditability, and deployment complexity. Full findings inside.
Research: Human-in-the-Loop Approval Latency in Agent Systems
15RL measures the impact of human-in-the-loop approval on AI agent throughput and user experience, identifying optimal approval strategies by risk category.
Research: Forensic Analysis of AI Agent Audit Logs
15RL develops forensic analysis techniques for AI agent audit logs, demonstrating how structured logs enable incident reconstruction and root cause analysis.
Research: Budget Control Mechanisms for AI Agent Spending
15RL evaluates budget control mechanisms for AI agent spending, analyzing per-session caps, token tracking, and cost attribution across agent architectures.
Research: Safety Metrics for AI Code Generation Agents
15RL proposes a standardized set of safety metrics for evaluating AI code generation agents, covering vulnerability introduction, secret leakage, and scope creep.
Research: Regulatory Compliance Requirements for AI Agents
15RL maps EU AI Act, SOC 2, and GDPR requirements to AI agent safety controls. Practical compliance checklist for teams deploying autonomous agents.
Research: Container Escape Risks in AI Agent Sandboxes
15RL evaluates container escape risks in AI agent sandboxing environments, testing Docker, gVisor, and Firecracker isolation against agent-driven exploits.
15RL Study: How AI Agents Expose Credentials
15RL documents how AI agents inadvertently expose credentials through logs, tool calls, and generated code, with data from 300+ agent session analyses.
Research: Cross-Agent Contamination in Multi-Tenant Systems
15RL documents cross-agent contamination risks in multi-tenant AI agent systems, where data and behavior from one agent session leaks into another.
Research Brief: Why Deny-by-Default Outperforms Allow-by-Default
15RL research shows deny-by-default policies reduce AI agent safety incidents by 94% compared to allow-by-default. Data from 3,000 deployment hours.
15RL Developer Survey: Attitudes Toward AI Agent Safety
Survey of 482 developers reveals attitudes toward AI agent safety tools. 73% say safety is important but only 22% have implemented controls. Full results.
Research: Risk Assessment for AI Agents in DevOps Pipelines
15RL assesses risks of deploying AI agents in DevOps pipelines, covering infrastructure-as-code manipulation, CI/CD compromise, and supply chain threats.
Research: AI Agent Safety in Educational Environments
15RL examines AI agent safety risks unique to educational settings, including student data protection, academic integrity, and content appropriateness controls.
Research: Barriers to Enterprise AI Agent Adoption
15RL identifies the top barriers preventing enterprise adoption of AI agents, with security and compliance concerns ranking above cost and technical complexity.
Research: File System Attack Vectors in AI Agent Deployments
15RL research identifies critical file system attack vectors in AI agent deployments and evaluates mitigation strategies for path traversal and data exfiltration.
Research: Compliance Requirements for Financial AI Agents
15RL maps financial regulatory requirements to AI agent safety controls, covering SOX, PCI-DSS, and emerging AI-specific financial regulations.
15RL Outlook: The Future of AI Agent Safety
15RL projects the future trajectory of AI agent safety, identifying emerging challenges, promising approaches, and the research priorities for the next 3 years.
Research: Government Standards for AI Agent Deployment
15RL analyzes government standards and frameworks applicable to AI agent deployment, including NIST AI RMF, FedRAMP, and executive orders on AI safety.
Research: Hash-Chain Audit Logs for AI Agent Accountability
15RL analyzes why traditional logging fails for AI agents and how hash-chain audit logs provide tamper-evident accountability. Technical evaluation inside.
Research: AI Agent Risks in Healthcare Applications
15RL identifies and categorizes risks specific to AI agent deployments in healthcare, from PHI exposure to clinical decision support failures.
15RL Framework: Incident Response for AI Agent Failures
15RL provides an incident response framework tailored to AI agent failures, covering detection, containment, analysis, and recovery procedures.
Research: Comparing Safety Features Across LLM Providers
15RL compares built-in safety features across major LLM providers for agent use cases, evaluating tool-call controls, rate limits, and abuse prevention.
15RL Audit: Security of Model Context Protocol Servers
15RL audits the security posture of Model Context Protocol servers, identifying authentication gaps, injection risks, and tool definition vulnerabilities.
15RL Guide: Minimum Viable Safety for AI Agents
15RL defines the minimum viable safety controls every AI agent deployment needs, providing a practical starting point for teams new to agent safety.
Research: Safety Challenges in Multi-Agent AI Systems
15RL analyzes safety challenges in multi-agent AI systems: trust propagation, shared resource gating, and privilege escalation across agent boundaries.
Research: Network Exfiltration Patterns in Unprotected AI Agents
15RL documents network exfiltration patterns observed in unprotected AI agents, including DNS tunneling, encoded payloads, and covert channel techniques.
Open Source vs Proprietary AI Safety: A Research Comparison
15RL compares open-source and proprietary AI agent safety tools across transparency, auditability, cost, and community trust. Open source wins on key metrics.
Research: Design Patterns for AI Agent Policy Engines
15RL analyzes five design patterns for AI agent policy engines: rule-based, LLM-as-judge, hybrid, capability-based, and graph-based. Technical comparison.
Research: Comparing Policy Languages for AI Agent Governance
15RL compares policy language approaches for governing AI agent behavior, evaluating declarative, imperative, and hybrid models across usability and expressiveness.
Research: Prompt Injection Impact on Tool-Using AI Agents
15RL research shows tool-using AI agents are 4.7x more vulnerable to prompt injection than chatbots. Action gating is essential where output filtering fails.
Research: Rate Limiting Strategies for AI Agent API Calls
15RL evaluates rate limiting strategies for AI agent API calls, comparing fixed window, sliding window, token bucket, and adaptive approaches.
15RL Case Studies: Real-World AI Agent Failures
15RL documents five real-world AI agent failure cases, analyzing root causes, impact, and lessons learned for improving agent safety practices.
15RL Recommended: The Minimal AI Agent Safety Stack
15 Research Lab recommends a three-layer AI agent safety stack: SafeClaw for action gating, container isolation, and runtime monitoring. Full architecture.
Research: AI Agent Safety in Regulated Industries
15RL examines how AI agent safety requirements differ in regulated industries, mapping compliance obligations to specific technical controls and audit needs.
Research: Runtime vs Static Analysis for AI Agent Safety
15RL compares runtime and static analysis approaches to AI agent safety, measuring detection rates, latency impact, and coverage across different threat types.
15RL Checklist: AI Agent Safety for Engineering Teams
15RL provides a practical safety checklist for engineering teams deploying AI agents, covering pre-deployment, runtime, and ongoing maintenance requirements.
15RL Methodology: Testing AI Agent Safety Policies
15RL presents a systematic methodology for testing AI agent safety policies, including red-team scenarios, regression testing, and coverage analysis.
Research: Effectiveness of Secrets Detection in AI Agent Pipelines
15RL measures the effectiveness of secrets detection tools in AI agent pipelines, finding that traditional scanners miss 31% of agent-exposed credentials.
Research: Shell Injection Risks Through AI Agent Tool Calls
15RL analyzes shell injection risks in AI agent tool calls, documenting how agents can be manipulated into executing arbitrary system commands.
15RL Survey: AI Safety Practices in Startups
15RL surveys AI safety practices in 89 startups deploying AI agents, revealing that speed-to-market consistently outweighs safety investment in early stages.
State of AI Agent Security 2026: 15RL Research Findings
15 Research Lab's annual overview of AI agent security in 2026. Adoption trends, threat landscape, tooling gaps, and recommendations for the year ahead.
Research: Reliability of Webhook Notifications in Safety Systems
15RL measures webhook notification reliability in AI agent safety systems, finding that delivery failures create blind spots in human oversight workflows.
Research: Workspace Isolation Techniques for AI Agents
15RL evaluates workspace isolation techniques for AI agents, comparing directory scoping, virtual filesystems, and ephemeral environments for security.
Research: Security Benefits of Zero-Dependency Architectures
15RL quantifies the security benefits of zero-dependency architectures for AI agent safety tools, analyzing supply chain risk reduction and audit simplification.