Versatile AI Risk Assessment
Knowledge baseAnalyses

Placing your own systemSystem types ThreatsSVG

The catalogue seen from the perspective of a concrete system type

Choose the system type and the architectural characteristics that apply. The grid shows, for each of the 52 threats, whether it is always to consider for this profile, needs review in the individual case, or remains atypical for this system type. A characteristic you set can put an atypical threat back up for review.

For architecture, product ownership and anyone assessing a concrete system

1. System type
2. Architectural characteristics

Systems with tool use, memory, planning steps, or autonomous execution need a stronger focus on agency, tool safety, session context, and human oversight.

37 Always consider Regularly relevant for this system type. The assessment should address this threat explicitly.
15 To review Relevant depending on architecture and deployment. The person carrying out the assessment decides case by case.
0 Atypical Atypical for this system type. This does not rule the threat out; matching architectural characteristics put it back up for review.

All 52 threats, bundled by topic group.

  1. Supply Chain and Provenance2/3

  2. Model and Training Data Manipulation1/4

  3. Prompt Attacks and Guardrail Evasion4/5

  4. Attacks on the Running Model and Service5/7

  5. Privacy and Data Leakage4/4

  6. Application and Integration Security7/7

  7. Harmful Content2/9

  8. Malicious Use for Attacks, Fraud and Disinformation3/4

  9. Agentic and Autonomous AI5/5

  10. Reliability and Responsible Use4/4

Review questions for this system type
  • Which tools can the agent invoke and with which privileges?
  • Which actions require explicit human approval?
  • Can external or stored context influence future agent decisions?
The file shows the current state including any filters and carries the note, catalogue version and source.

The assignment is a working basis for the assessment, not a clearance. “Atypical for this profile” does not mean “ruled out”: the professional decision still rests with the person carrying out the assessment.

Versatile AI Risk Assessment is an aid for structuring AI risks and making them transparent. It does not replace legal or professional advice and makes no binding decisions.

Further analyses

The same data, a different question

The other nine analyses, searchable and filterable by perspective like the overview; those of the same perspective first.

Show filters
No filter set, the list shows the complete collection.9 of 9 further analyses

Understanding threats

Lifecycle

When does which risk arise?

Planning mitigations

Effect matrix

How and where do mitigations take effect?

Planning mitigations

Ownership

What types of mitigations are there?

Checking frameworks and evidence

Coverage gaps

What does my framework not reach?

Checking frameworks and evidence

Evidence base

What is this based on?

Checking frameworks and evidence

Framework mappings

Where is this anchored professionally and legally?

Overview of all ten analyses with all filters

Related
Threat catalogue

All 52 threats in full: description, impact, example, mitigations and verified sources.

To the catalogue
EU AI Act quick check

Six questions along the EU AI Act and a deadline plan with the obligations that apply to your system.

To the quick check
Obligations compass

Choose a role and a risk class and see which EU AI Act obligations apply from when, with legal reference and evidence.

To the obligations compass
Assess it yourself

The full assessment with system context, threat selection and evidence tracking, free of charge in the browser.

To the live demo