Threat modelling AI applications
Data flows, trust boundaries and tools of an AI application are mapped to identify where untrusted inputs could manipulate the model or where data could leak.
GCRI · Artificial Intelligence and Autonomous Systems
Cybersecurity and cryptography protect AI systems and the data they use. In this field the focus is threats specific to AI, such as prompt injection, poisoning and model theft, alongside conventional security of the software and infrastructure around models.
Cybersecurity and cryptography protect information systems and data from unauthorised access, disruption and manipulation. They address threats that can spread rapidly across interconnected digital and physical systems. In artificial intelligence and data and compute, cybersecurity analysis identifies how a single compromise can propagate across sectors that share software, vendors or networks. It informs resilience requirements and cyber insurance. The canonical description is on the STEM Commons.
Data flows, trust boundaries and tools of an AI application are mapped to identify where untrusted inputs could manipulate the model or where data could leak.
Provenance of datasets and model weights is recorded and verified with signatures and hashes before use.
Applications are tested with suites of injection attempts, and the success rate is tracked as defences change.
The screen a team would use to follow this method: its measures, states and decisions. Names are examples and values are illustrative.
AI and autonomous systems · Cybersecurity for AI systems · Run R-5953
| Weak key rotation | Identity service | Phezulu | PR.AA | Vula |
| Unpatched gateway | Edge | Phezulu | PR.PS | Fixing |
| Logging gap | Payments API | Medium | DE.CM | Vula |
| Stale accounts | Directory | Phansi | PR.AA | Closed |
Datasets and records used by cybersecurity and cryptography on this platform, with the role each plays and the label its values carry. Sources are listed in the source register.
| Idatha | Indima | Label |
|---|---|---|
| MITRE ATLAS techniques | Threat catalogue | REPORTED |
| Vulnerability databases | Known flaws | REPORTED |
| Security test results | Defence performance | OBSERVED |
A hazard footprint takes out co-located assets in every layer; fragmentation, dependency between layers and overload of the survivors then spread the failure. The 3D view separates the layers by depth. Teaching model with stated illustrative values, not a prediction.
Controls
Readouts
| Step | Data centres working (%) | Networks working (%) | Power working (%) |
|---|---|---|---|
| 0 | 94 | 100 | 100 |
| 1 | 89 | 67 | 100 |
| 2 | 72 | 67 | 100 |
| 3 | 72 | 67 | 78 |
The model runs on the teaching values set by the controls. To calibrate it for a place, read it against these admitted layers and series, which this page’s map can draw:
Across the six pillars, led by Nexus Consortium
Controls are tested through red-team exercises, audits and formal analysis of cryptographic schemes. Incident data are used to check whether defences performed as intended.
| Check | How it is done |
|---|---|
| Threat model review | Threat models are reviewed when the application or its tools change |
| Provenance verification | Model artefacts are checked against signed records |
| Regression testing | Security tests are rerun after each update |
Adversaries adapt, so historical data may not reflect future attacks. Software supply chains and legacy systems introduce weaknesses that are hard to inventory.
The model, its assumptions, its data and its uncertainty, in the form a scientist or engineer would check. Structures follow published methods. The values drawn are illustrative, not findings.
The model structure used for Cybersecurity for AI systems, the assumptions it makes and how each is tested.
Ri = f(Hi, Ei, Vi, Ci)hazard, exposure, vulnerability and coping capacityEAL = ∑k pk · Lkover events k with annual probability pkxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error| Assumption | Why it is made | How it is tested |
|---|---|---|
| Hazard and exposure are independent within an event | Keeps the loss integral tractable | Compare with a copula fit on joint records |
| Couplings are linear near the operating point | Lets shocks propagate through A | Perturb beyond the linear range and compare |
| Vulnerability curves transfer between places | Local curves are rarely available | Validate against observed losses where they exist |
Cybersecurity and Digital Resilience stewards cybersecurity and cryptography for the Consortium and reviews how it is applied across platforms.
Reports is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.
Ungalokothi uthumele idatha yomuntu siqu, imininingwane ebucayi yengqalasizinda noma noma yini ongenalo ilungelo lokuyabelana ngayo. Umnikelo ubuyekezwa ngaphambi kokusetshenziswa; ukuwuthumela akunikezi ubuqambi noma amalungelo okuhlela, futhi ukusekela akuthengi muphumela.
Information and enquiries are open today. An enquiry receives a reference and names the next step; applications, payments, compute allocations and registrations open only when their team, terms and review capacity exist.
Maps, data and figures on this site are shown for context and general information. They are not advice, forecasts or warnings; the limits of the information are set out in the Disclaimers and Data Notices.
This site uses only strictly necessary cookies to work and to remember your privacy choices. Nothing loads from other companies, and we never sell or share your personal information.
This site uses only strictly necessary cookies to work and to remember your privacy choices. Nothing loads from other companies, and we never sell or share your personal information. We run no advertising. You can review your choices at any time through Privacy choices at the foot of every page.