Pentesting & Recon

Test LLM apps and agents for prompt injection, tool abuse, RAG poisoning and MCP risks

Map an attack surface with passive and active recon, subdomains, ports and fingerprints

Chain low-severity primitives into exploit paths when no single bug gives code execution

Test cloud metadata, IAM privilege escalation and Kubernetes exposure during pentests

Research CVEs, public exploits and advisories for a component before attempting an exploit

Test initial access with phishing, password spraying, AiTM and device code flows

Run a full-chain pentest by building a fact graph and deriving attack paths from evidence

Keep a cross-session fact graph of pentest findings, evidence and vulnerability records

Format pentest findings, vulnerability reports and change logs in a consistent structure

Require evidence before a finding is marked confirmed and log negative results

Cover privilege escalation, lateral movement, tunneling and password cracking in tests

Switch proxies and bootstrap pentest tools when requests hit 403, 429 or WAF blocks

Apply red team opsec for stealth, rate timing, traffic blending and minimal footprint

Apply specialized exploitation playbooks for GoEdge, ARP MITM, BT Panel and MinIO

Set the broad pentest mindset for multi-domain testing within the engagement scope

Test web apps for injection, SSRF, auth bypass and WAF or CDN evasion techniques

Test WiFi, BLE, RFID, SDR and hardware interfaces with side-channel and fault methods

Find zero-days with variant analysis, patch diffing, fuzzing and taint tracing

Search Burp Suite project files for headers, bodies, findings and proxy history

Scan Android APKs for open Firebase databases, buckets and auth misconfigurations

Run authorized red-team reviews of AI agents, MCP servers and skills with reports

Probe an AI agent for privilege escalation and unauthorized access through dialogue

Find error chains and single points of failure that cascade through agent workflows

Probe an AI agent for leaks of system prompts, API keys and personal data

Test an AI agent for direct prompt injection and system role override attempts

Test whether an AI agent induces unsafe user actions through deceptive replies

Test whether hidden instructions in documents, RAG or web content hijack an agent

Check messages between agents for data leaks and privilege mismatches

Test whether an agent stores hostile instructions in long-term memory

Test agent memory and RAG knowledge bases for poisoned or contaminated content

Test whether an agent misuses file, code execution or network tools

Find command injection, eval misuse and arbitrary code loading in agent systems

Test whether an agent can leak user data through URL paths and chained fetches

Test LLM apps for direct and indirect prompt injection and prompt leaks

Test LLM guardrails against translation, encoding and fictional-framing evasion

Test LLM system prompts against override, constraint negation and role reversal

Test multi-turn memory poisoning and cross-session context leaks in LLM apps

Test LLM APIs for rate-limit abuse, cost amplification and schema bypass

Assess LLM training and feedback pipelines for data and reward poisoning

Test LLM outputs for covert channels, schema breaks and data exfiltration

Test LLM deployments for misinformation, impersonation and hallucination abuse

Test multimodal LLM inputs for hidden text and image or audio injection

Test LLM apps for system prompt, training data and key extraction risks

Test agent tool chains for MCP tool poisoning and agent-to-agent injection

Test RAG pipelines for knowledge base poisoning and retrieval manipulation

Assess AI model and dataset supply chains for malicious artifacts

Test LLM services for denial of service, quota exhaustion and billing abuse

Assess human-facing AI risks such as deepfake and voice clone social engineering

Write abort and crisis plans with halt triggers and response actions