AI OFFENSIVE SECURITY / RED TEAM RESEARCH
Real security boundaries for AI that takes action.
Chengdu Yofune Ariake Technology Co., Ltd. provides security products, professional services and tailored solutions, with AI security at its core. From models and agents to business systems and infrastructure, we bring offensive research into real business environments.
As AI moves from understanding to action, security needs clearer boundaries. Drawing on research from Yofune_LABS, we use reviewable attack evidence to uncover risks, guide remediation and give every improvement a basis in evidence.
Cross the boundary. Verify the behavior.
We study how AI can be influenced, how it makes decisions, and how it changes the real world. Following an attack path reveals the questions a system needs to answer.
The attack surfaces we study
THE HYPOTHESIS
Trace behavior after untrusted content enters the context. Connect inputs, decisions and tool calls to establish whether the impact is real.
Input sample / Call trace / Impact
Agree on environment, goals and boundaries.
Test risk hypotheses along real paths.
Connect inputs, execution and conclusions.
Verify the change along the same path.
SECURITY SERVICES
AI red teaming and offensive research are our foundation. We also work across enterprise networks, cloud environments and business systems: from architecture and assessment to operations and training, with agreed scope, reviewable evidence and practical deliverables.
Design around business needs and assets, connecting existing systems so policies, identities and logs remain manageable.
Discuss your requirementsMap network zones, access controls and logging paths. Support security product selection, deployment, policy migration and integration testing with clear system responsibilities.
Identify technical and management gaps against applicable Chinese MLPS requirements. Support implementation planning, remediation and documentation in coordination with your team and assessors.
Plan least-privilege access, secrets management, image checks and audit logging across cloud accounts, networks, containers and delivery pipelines.
Map sensitive data flows and user and service permissions. Integrate authentication, access auditing, data masking and outbound data controls.
Map industrial assets, network separation and remote maintenance paths around production continuity requirements. Agree change windows and validation methods with operations teams.
Connect risk discovery with attack evidence, remediation priorities and retesting to establish whether controls have improved.
Discuss your requirementsReview assets, business dependencies, exposed services and existing controls. Combine interviews and technical checks to distinguish business impact and treatment priorities.
Test web applications, APIs, mobile applications and internal networks within authorized scope. Validate identity, authorization and business logic issues with reviewable attack evidence.
Combine manual analysis and tooling to trace inputs, sensitive operations and authorization checks. Review vulnerable dependencies, exposed secrets and critical business logic.
Assess LLMs, RAG applications, AI agents and MCP toolchains for prompt injection, sensitive data exposure, excessive tool permissions and cross-component trust failures.
Review accounts, permissions, services and logging across hosts, databases, middleware and network devices. Apply agreed hardening changes and check their business impact.
Use safe test samples in authorized mailboxes and agreed scope to assess impersonation detection, URL and attachment blocking, quarantine, alerts and response workflows.
Support recurring security work across alerts, vulnerabilities, incidents and critical periods, with actionable records and clear handovers.
Discuss your requirementsReview agreed logs and security alerts, establish context and reduce noise. Coordinate response through jointly defined severity and escalation procedures.
Plan reviews around asset changes and newly disclosed vulnerabilities. Verify exploitation conditions and impact, then track owners, remediation status and retesting.
Help investigate intrusions, malware and compromised accounts. Preserve relevant logs and evidence, assess impact, recommend containment and validate recovery.
Prepare for major events or critical business windows through risk reviews, policy checks and rehearsals. Agree coverage periods, coordination procedures and escalation contacts.
Review security device health, policy changes, account permissions and log collection. Help retire outdated rules and confirm the effect of configuration changes.
Build training and exercises for employees, developers and security teams around the decisions they face in everyday work.
Discuss your requirementsCover email impersonation, account protection, data handling and AI tool use. Use authorized simulations to evaluate recognition and reporting procedures.
Teach authentication, authorization, input handling and dependency management in the team's technology stack. Use code exercises to practice finding, fixing and validating issues.
Help AI product and engineering teams understand prompt injection, retrieval poisoning, tool permissions and sensitive data risks through controlled validation and defense exercises.
Run tabletop exercises or isolated practical labs around agreed incident scenarios. Practice analysis, escalation, containment, recovery, communication and handover.
Start with defined questions and reproducible experiments, then turn offensive research into assessment methods, defensive guidance and useful engineering work.
Discuss your requirementsInvestigate model context, RAG data, agent memory and MCP toolchains. Test risk hypotheses, attack prerequisites, propagation paths and the limits of controls.
Analyze agreed products or components for vulnerability causes, exploitation prerequisites and affected versions. Reproduce issues in controlled environments and evaluate fixes.
Connect risks across identity, networks, applications and data. Use authorized comparative experiments to establish when detection, isolation and access controls are effective.
Develop test scripts, rules and tool prototypes for defined assessment or operations needs, with explicit inputs, outputs, applicability and reproduction steps.
Security products, built from research
Select a panel to explore the product
Connect task planning, tool collaboration, attack-path reasoning and evidence into controlled, agentic security assessments of complex systems.
Interactive product demo · Sample data
Use tool results and asset context to organize the next validation steps.
Researchers. Engineers. Red teamers.
Our founding members come from leading security companies including Chaitin, Topsec, 360 and Qi-Anxin, with an average of over seven years of professional experience across red team operations, vulnerability research and security product engineering.
Yofune_LABS is Yofune's offensive security research lab. We take an attacker's perspective: what does a system trust, what can an agent do, and can that boundary be crossed? From model context to tool execution, from vulnerability research to product engineering, we turn hypotheses into controlled validation, then bring that research into products, services and practical methods.
SECURE THE UNKNOWN
BEFORE WE BEGIN
We assess LLM applications, RAG knowledge bases, AI agents and toolchains for prompt injection, sensitive-data exposure, permission boundaries and unintended execution, and we trace the path through application code and business flow.
Yes. With AI security at our core, we also offer scoped risk assessments, penetration testing, code reviews, security baseline checks, security operations and incident response, training, and email gateway security assessments. Targets, authorization boundaries, deliverables and retesting are agreed before work starts.
Yes. Deployment and scoped assessments can follow your data and hosting constraints. Scope, model access, data flow and deliverables are agreed before work starts.
A reviewable risk conclusion: impact, paths, evidence and remediation. Retests and follow-on assessments can be scheduled from that same scope.
LET’S QUESTION WHAT’S NEXT.
Tell us about your AI application, agent or business environment. Let’s define the right scope together.