What’s at stake
The environments we support operate some of the most power-dense and operationally sensitive infrastructure in the world.
A security incident is not just an IT problem. It can impact availability, operational efficiency, and the systems responsible for supporting large-scale AI workloads. Phaidra’s security architecture is designed to minimize operational risk while preserving the reliability and continuity our customers depend on.
Two products, two postures
Phaidra offers a portfolio of products with different operational risk profiles. We're open about which is which, so you can choose where to start and how far to go.
Intelligence & Observability
Prism is where most customers begin. It ingests telemetry from your systems, contextualizes it with site-specific documentation, and surfaces operational insights to your team.
Prism is read-only by design. It has no write paths into your control systems and cannot make changes to your environment.
Agentic
Beyond Prism, we offer a portfolio of agents, starting with the Liquid Cooling Agent, that take action on specific control problems. Where Prism observes, the AI agents act.
Security through constrained autonomy
Every Phaidra agent is designed around a single objective and a defined set of permissions. Rather than creating general-purpose AI with broad authority, we deploy narrowly scoped agents responsible for specific control problems and equipment classes.
This approach reduces blast radius, simplifies security review, and creates clear authorization boundaries between systems. The result is a platform that can scale operational capabilities without scaling operational risk.
Customer-controlled connectivity
Phaidra connects to your environment on your terms, not ours.
- Data is pushed outbound from infrastructure you control
- Phaidra does not initiate inbound connections
- A single outbound TLS connection is sufficient
- No persistent remote access is maintained
- Updates are reviewed and deployed by your team
This architecture allows customers to adopt Phaidra while maintaining existing network segmentation and security controls.
Governance controls for AI
AI systems operating in critical environments require clear controls around data usage, retention, and model lifecycle management.
Phaidra provides:
- No training on customer operational data
- Regional data residency
- Defined retention and deletion policies
- Customer notification for underlying model changes
These controls help customers meet internal governance requirements while maintaining visibility into how AI capabilities are deployed and managed.
Security controls
- Production access restricted by default and granted only for approved business use
- Hardware-backed MFA for all employees
- Infrastructure-as-code with peer review and two-person approval requirements
- Continuous threat modeling and security testing
- Third-party penetration testing across application, runtime, and infrastructure layers
- Full audit trails for production access
Independently verified
We hold our security program to standards verified by independent auditors.
SOC 2 Type II
AICPA standards. Annual audit with interim bridge letter.
ISO/IEC 27001:2022
Certified through ANSI National Accreditation Board.
ISO/IEC 42001 (roadmap)
Expanding our assurance program to cover AI management systems.
Reports, certificates, and supporting documentation are available through our Trust Center under NDA
AI Transparency & explainability
Phaidra is designed to provide visibility into:
- What the AI is doing
- Historical system behavior
- Planned operational actions
We believe transparency and explainability are critical for building trust in AI systems operating within industrial and mission-critical environments.
Our approach focuses on turning AI systems from “black boxes” into understandable and observable operational systems for plant operators and infrastructure teams.