Available for mandates

Finding the gaps compliance reviews miss.

I apply forensic investigation methodology to deployed agentic AI — finding the behavioural control gaps your compliance framework cannot reach, and producing the evidence that holds up when a regulator or underwriter asks the hard questions.

Get in touch → About →
Insurance Underwriters Regulated Institutions Legal Counsel Workshops · Interim Mandates EU AI Act · FINMA 08/2024
Latest Analysis
May 2026 · AI Governance

What Your Agent Remembers — and Who Governs That

AI agents accumulate memory across sessions — and that memory changes how they behave over time. Most governance programs have not drawn the boundary around it.

Read full analysis →
01 Approach

Eight years in forensic and cyber investigation. One discipline above all: find the structural gap, not the actor. That is the methodology behind the Agentic Control Plane.

8 yrs forensic & cyber investigation · FINMA · EU AI Act · ISO 42001 · NIST AI RMF · Full background →
02 Explore My Research
Aljona Schwan — Founder, AI Resilience Lab

Author

Aljona Schwan

Founder · AI Resilience Lab · Zürich

Writing on agentic AI governance, model risk, and the structural control failures that emerge when autonomous systems operate without adequate behavioural oversight. Grounded in primary sources, regulatory frameworks, and eight years of forensic and cyber investigation practice.

LinkedIn All Research
May 2026 AI Governance
What Your Agent Remembers — and Who Governs That.
AI agents accumulate memory across sessions — and that memory changes how they behave over time. Most governance programs have not drawn the boundary around it.
Read →
Apr 2026 Data Protection
The Verification Layer.
Biometric proof-of-humanity systems are emerging as a response to deepfake fraud. The governance architecture underneath the current implementations has not caught up.
Read →
Apr 2026 AI Security
The Mythos Reckoning.
Autonomous vulnerability research has arrived. What a 27-year-old OpenBSD bug and a lunchtime email tell us about the structural limits of human-speed defence.
Read →
Apr 2026 Agentic AI Risk
Agent Intervention Taxonomy.
A governance framework for organisations deploying autonomous AI agents in production — structured across five intervention pillars with implementation prioritisation for Risk, Control, and Board-level oversight.
Read →
Apr 2026 Forensic Lens
The monitoring gap NIST cannot close.
NIST AI 800-4 documents deceptive AI behavior as an unsolved security monitoring problem. Pre-deployment evaluations exist — but face a structural ceiling. What that means for agentic deployments in regulated industries.
Read →
Ongoing All Writing
More on AI control failure, behavioural governance, and agentic AI risk.
Long-form analysis on the governance gaps that compliance reviews miss and that adversarial actors exploit first.
All writing →
03 What I Offer

Engagements scoped to a defined output.

From building capability inside an institution to assessing what is already deployed — the engagements are ordered by accessibility and immediate relevance. Every engagement applies forensic investigation methodology to the same question: what is this agentic system actually doing, by what route, and under whose authority?

View full engagement detail →
01

Agentic Control Failure Workshop

Half-day practitioner curriculum for risk, compliance, audit, and security teams — fixed curriculum, fixed price, virtual-first, global delivery in English and German. The fastest entry point into understanding what agentic control failure looks like in a live production environment and what to do about it.

  • AI Control Failure Taxonomy · five failure modes · incident-based case analysis
  • EU AI Act and FINMA regulatory obligation map for deployed agents
  • Institutional booking or open enrolment · half-day · English and German
02

Interim AI Risk Management

Embedded AI risk management leadership for regulated financial institutions navigating FINMA Guidance 08/2024 and EU AI Act deployer obligations before a permanent hire is in place. Three to six months, 20–30% FTE — building the function from the inside and producing a governance structure the institution owns permanently.

  • EU AI Act August 2026 deadlines · FINMA 08/2024 · embedded practitioner model
  • Functioning AI risk framework · Agentic Control Plane design · regulatory evidence package
  • Documented handover to permanent function or ongoing advisory retainer
03

Agentic Governance Advisory

Fixed-fee engagements for regulated financial institutions. AI control framework design, Agentic Control Plane architecture, EU AI Act and FINMA readiness — translating governance commitments into the control structures a regulator can audit and a risk committee can act on.

  • FINMA Guidance 08/2024 · EU AI Act Articles 9, 14, 26
  • Agentic Control Plane design and implementation
  • Fixed-scope or phased engagement, scoped to the deployment
04

Specialist Assessments & Forensic Review

Three specialist engagements for specific buyer contexts: Agentic Control Assessment for security leadership who need a behavioural capability map of their deployed agent; AI Agent Control Assessment for Underwriting for carriers and insured organisations pricing agentic AI risk; and Forensic Second Opinion for institutions whose prior AI governance assessment did not reach the execution layer.

  • Agentic Control Assessment — production behavioural assessment · CISO buyer
  • Underwriting Assessment — risk score · carrier summary · coverage due diligence
  • Forensic Second Opinion — examiner-ready documentation · execution layer gap analysis
04 Contact

Get in touch

Each engagement begins with a discovery call to understand the deployment, the regulatory context, and the gap that needs closing. The proposal that follows is scoped to the deployment, not to a generic framework.

Available for workshops, interim AI risk management mandates, governance advisory, agentic control assessments, underwriting assessments, forensic second opinions, and speaking engagements. Delivered remote-first, with on-site working sessions where the deployment requires it.