Insights
Analysis for Governing High-Risk AI
Practical analysis, industry intelligence, and original perspectives on runtime governance, accountable authority, AI assurance, and the emerging control requirements of consequential AI.
01 / Featured Technical Perspective
Your AI Audit Logs Are Fiction
An AI audit log can tell a convincing story. But can it withstand independent scrutiny?
Piscys examines decision-time evidence, tamper-evident records, and what it takes to make AI governance independently verifiable.
by Jeff Hertzig · CTO / Founder · October 21, 2026
02 / Founder Insights
Piscys Perspectives on Governing
High-Risk AI
Original thinking on AI governance, assurance, trust, and the operational and human questions shaping how consequential systems should be designed, controlled, and held accountable.
An AI audit log can tell a convincing story. But can it withstand independent scrutiny? Piscys CTO Jeff Hertzig explores decision-time evidence, tamper-evident records,
How training, feedback, and human behavior shape AI systems — and why safer intelligence begins upstream of deployment.
Why responsible AI requires more than principles: durable infrastructure for oversight, accountable authority, and consequential action.
03 / Explore by Topic
Explore the Questions Shaping AI Governance
Browse practical analysis, industry intelligence, and longer-form perspectives on the governance challenges emerging as AI becomes more capable and consequential.
Runtime Governance
Industry & Regulation
Financial Services & Healthcare
Perspectives
04 / What We're Watching
The Signals Reshaping Governance
Emerging developments in AI autonomy, regulation, and enterprise risk — and what they mean for operational control, accountable authority, and measurable assurance.
When AI Agents Cross the Boundary Into Real Systems
Anthropic investigated four incidents in which Claude accessed third-party systems during cybersecurity evaluations. Misconfigured environments exposed models to the live internet rather than their assigned simulations.
Why It Matters
Agent assumptions an action is permitted cannot substitute for enforceable authorization. Enterprises need independent boundaries, proportional intervention, and evidence of what actually occurred — not an agent’s account of its behavior.
AI Oversight Moves Toward Enforcement
The EU AI Board reviewed AI Act implementation, enforcement priorities, frontier-model capabilities, and recent AI incidents. The discussions signal continued coordination around how AI oversight translates into practice.
Why It Matters
Enterprise governance increasingly needs to connect applicable requirements to operational safeguards, authorized exceptions, and verifiable records. Policy commitments alone cannot demonstrate how consequential AI decisions were actually governed.
Runtime Controls Become Enterprise Infrastructure
HPE and NVIDIA announced integration plans combining governed agent execution, out-of-process policy enforcement, operator approvals, and auditable activity. OpenShell is generally available; parts of HPE are planned.
Why It Matters
Independent controls are moving into the enterprise AI infrastructure stack. The next question is how organizations coordinate enforcement, proportional intervention, decision authority, and defensible evidence across AI workflows.
05 / Piscys Research & Concepts
Defining the Measures and Mechanisms of AI Assurance
From measurable risk exposure to accountable decision authority and multi-constraint evaluation, Piscys is developing original concepts designed to make AI governance actionable, observable, and auditable.
Piscys research and architectural concepts are in development. Descriptions reflect proposed approaches, not validated product capabilities or demonstrated outcomes.
Exposure Minutes
A proposed operational metric measuring how long a material AI risk condition remains active before accountable containment—shifting assurance toward measurable exposure and response.
MAKE EXPOSURE MEASURABLE
Accountable Authority
A model for making decision rights explicit: who may establish constraints, authorize actions, approve overrides, and accept residual risk when AI systems act.
MAKE AUTHORITY EXPLICIT
Constraint Polyphony
A multi-lens approach to evaluating consequential AI actions against interacting governance constraints, informing proportional intervention, accountable authority, and supporting evidence.
MAKE CONSTRAINTS OPERABLE
06 / From Insight to Action
Bring the Discussion Into Your AI Environment
If your organization is deploying AI into consequential workflows, Piscys can help identify where governance relies on documentation — and where runtime controls, accountable authority, or stronger evidence may be needed.