Results for “prompt-injection”

Audit installed SKILL.md files and companion scripts for shell injection, secret exfiltration, path traversal, prompt-override payloads, destructive commands, and obfuscation. Integrates with Bankr Safety Scores and produces delta reports against prior scans.

Wire Promptfoo and DeepTeam into CI/CD for automated regression red-teaming of LLM apps against OWASP LLM Top 10 and OWASP Agentic presets, failing the build when jailbreak or injection vulnerabilities regress.

Test LLM applications for system prompt leakage using manual payloads, garak, and Promptfoo to extract embedded secrets and routing logic.

Analyzes prompts for safety, bias, security vulnerabilities, and effectiveness, providing detailed improvement recommendations with frameworks, testing methodologies, and educational content.

Assess AI/ML systems for prompt injection, jailbreak vulnerabilities, model inversion risk, data poisoning exposure, and agent tool abuse, with MITRE ATLAS mapping and guardrail recommendations.

Assigns behavioral trust grades (A–F) to MCP servers by running probes for prompt injection, permission overreach, data leaks, and adversarial-input handling, and publishes reproducible onchain attestations.

Deploy Llama Guard, NeMo Guardrails, and LLM Guard as runtime input/output scanners to block jailbreaks, prompt injection, and toxic content in production LLM applications.

Conduct authorized security assessments of LLM applications and AI agents, covering prompt injection, tool abuse, RAG exposure, memory poisoning, and model supply-chain risks.

Audit Claude Code configuration files for security vulnerabilities, misconfigurations, and injection risks using AgentShield.

Run NVIDIA garak probe suites against an LLM endpoint to test for jailbreaks, prompt injection, data leakage, and toxic generation, then interpret the hit-rate report for triage and reporting.

Builds input and output validation guardrails for LLM-powered applications to prevent prompt injection, data leakage, toxic content generation, and hallucinated outputs using NeMo Guardrails, Presidio, and Guardrails AI.

Authorized OT/ICS security assessment covering Purdue model zoning, PLC/SCADA exposure, industrial protocol discovery, and safe passive-first evaluation.

Scans AI agents, MCP servers, and skills for security vulnerabilities from the command line, detecting prompt injections, tool poisoning, toxic flows, malware payloads, and credential handling issues across 15+ risk categories.

Audits Claude Code configuration files for security vulnerabilities, misconfigurations, and injection risks using AgentShield, covering CLAUDE.md, settings.json, MCP servers, hooks, and agent definitions.

Bypasses safety filters on API-served LLMs using jailbreak templates, input obfuscation, and multi-model racing.

Specialized CTF sandbox skill for analyzing prompt-to-tool chains, deployment drift, and supply-chain provenance in cloud, container, and CI/CD environments.

Blocks dangerous shell commands and warns on risky ones, applying a safety protocol before executing any command.

Apply least-privilege tool allowlisting, identity binding, and human-in-the-loop controls for agent tool calls.

Scan and audit AI agent skills for security risks before installation, producing a PASS/WARN/FAIL verdict with findings and remediation guidance.

Scans local repos and AI agent skill folders for security drift, saves trusted baselines, and compares changes after updates to highlight risky new capabilities.

Store and retrieve API keys, passwords, and credentials in an HSM-backed encrypted vault via MCP or REST API, with policy-based access control, secret rotation, multi-chain signing, and prompt injection scanning.

Audits GitHub Actions workflows for security vulnerabilities in AI agent integrations, detecting attack vectors where attacker-controlled input reaches AI agents in CI/CD pipelines.

Automate multi-turn adversarial conversations against LLM agents using Microsoft PyRIT, including Crescendo and Tree-of-Attacks-with-Pruning (TAP) attack chains with scorer feedback loops.