AI Security & Agent Safety
AI Security & Agent SafetyMike leads the AI Security & Agent Safety track, focusing on how AI systems can be used safely inside real enterprise environments. His background combines enterprise architecture, daily software engineering, integration boundaries, automation, blockchain and smart-contract security, and governed AI execution. This track helps clients understand where AI systems can fail or be misused, what controls are needed before automation is expanded, and how security and safety become part of the implementation design from the beginning.
Mike also contributes to Yuauri’s senior review layer, together with Joni. This review layer helps ensure that technical recommendations, risk boundaries, governance assumptions, and client-facing conclusions remain consistent before they are presented.
Track ownership
Mike owns this specialist track as Yuauri’s AI security and agent-safety review area. The focus is to identify where AI-enabled workflows introduce new attack surfaces, unsafe permissions, data exposure risks, or uncontrolled automation paths — and translate those findings into practical implementation boundaries for client work.
Specialist focus
These areas describe the track owner's current specialist focus and the practical AI questions this track follows.
- Agent safety boundaries
- Prompt injection and tool misuse risks
- Permission and sandboxing patterns
- Data leakage and red-team scenarios
- Secure agent deployment
- Tool permission models
- Safety evaluation
- Misuse-resistant workflow design
Track-specific AI organization
For AI Security & Agent Safety, the support team is organized around failure modes, misuse paths, permission boundaries, and safe automation design. The AI-agent roles help inspect prompt-injection risks, tool access, data exposure, sandboxing assumptions, red-team scenarios, and safety evaluation needs before recommendations reach the client.
Track-specific AI organizations are adaptable. Custom agents can be added when a client situation requires a specialized role, workflow, control step, or evaluation function beyond the standard track support team.
Agents do designed work; the human track lead validates output and owns every client-facing recommendation.
What this track helps with
AI systems create new security boundaries. Prompts can carry hidden instructions, agents can call tools, retrieval systems can expose sensitive context, and automated workflows can act on behalf of users. This track helps clients identify these risks early, define safe permission boundaries, design approval paths, and test whether AI-enabled systems remain secure as prompts, tools, models, and workflows change.
Common client questions
- Can we safely let AI agents call tools or trigger workflows?
- What should AI never be allowed to do automatically?
- How do we prevent prompt injection and tool misuse?
- How do we design approval boundaries for agentic systems?
- What should be logged, reviewed, or audited?
- How do we test whether the system remains safe as prompts, tools, models, or workflows change?
Typical outputs
These outputs help a client decide where AI-enabled automation is safe, where human approval is required, and what controls should be in place before agents or AI workflows are expanded.
- Agent safety risk notes
- Prompt-injection threat model
- Tool permission recommendation
- Data leakage review checklist
- Sandbox boundary pattern
- Red-team scenario set
- Safety evaluation baseline
- Secure deployment backlog
Tools currently under observation
Named tools currently watched, tested, or validated within this track. Inclusion reflects active evaluation, not endorsement.
Published notes and evaluation fragments
Short technical write-ups connected to this track.