Skip to content

Commit 519e26f

Browse files
adding control lists to essentials step 1
1 parent 780f2fe commit 519e26f

1 file changed

Lines changed: 8 additions & 0 deletions

File tree

content/ai_exchange/content/docs/ai_security_overview.md

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -314,14 +314,22 @@ This section serves as THE starting point to understand the foundations of AI se
314314

315315
Note: Attackers that obtain the model (or a similar one) can perform or prepare input attacks efficiently and without being noticed.
316316

317+
---
317318

318319
**New controls** (overview [here](/go/controlsoverview/)):
319320

320321
The controls for AI security can be divided into four main categories and 8 sub categories, linking to over 50 individual controls discussed in depth at the Exchange.
321322

322323
Manage:
323324
- Extend existing [Governance](/go/aiprogram/), [Risk](/go/riskanalysis/) and [Compliance](/go/checkcompliance/) - in order to secure AI, you need overview, analysis, policy, training, and responsibilities
325+
> [AI PROGRAM](/go/aiprogram/ ), [DEV PROGRAM](/go/devprogram/), [CHECK COMPLIANCE](/go/checkcompliance/), [SEC EDUCATE](/go/seceducate/)
324326
- Extend existing **conventional security controls** to protect the AI-specific assets
327+
- Mananeging security:
328+
> [SEC PROGRAM](/go/secprogram/), [SECDEV PROGRAM](/go/secdevprogram/)
329+
- Development-time: model & data storage, model & data supply chain, data science documentation:
330+
> [DEV SECURITY](/go/devsecurity/), [SEGREGATE DATA](/go/segregatedata/), [DISCRETE](/go/discrete/)
331+
- Runtime: model storage, model use, augmentation data (including system prompts), and model input/output:
332+
> [MODEL ACCESS CONTROL](/go/modelaccesscontrol/), [RUNTIME MODEL INTEGRITY](/go/runtimemodelintegrity/), [RUNTIME MODEL IO INTEGRITY](/go/runtimemodeliointegrity/), [RUNTIME MODEL CONFIDENTIALITY](/go/runtimemodelconfidentiality/), [MODEL INPUT CONFIDENTIALITY](/go/modelinputconfidentiality/), [ENCODE MODEL OUTPUT](/go/encodemodeloutput/), [LIMIT RESOURCES](/go/limitresources/), [AUGMENTATION DATA CONFIDENTIALITY](/go/augmentationdataconfidentiality/), [AUGMENTATION DATA INTEGRITY](/go/augmentationdataintegrity/), > [CONF COMPUTE](/go/confcompute/), [MODEL OBFUSCATION](/go/modelobfuscation/)
325333
- Extend [Supply chain management](/go/supplychainmanage/) to incorporate suppliers providing data, models, and hosting
326334

327335
Have resilient models:

0 commit comments

Comments
 (0)