USA flag logo/image

An Official Website of the United States Government

Static Analysis of AI Systems

Award Information

Department of Defense
Defense Advanced Research Projects Agency
Award ID:
Program Year/Program:
2004 / SBIR
Agency Tracking Number:
Solicitation Year:
Solicitation Topic Code:
Solicitation Number:
Small Business Information
GrammaTech, Inc
531 Esty Street Ithaca, NY -
View profile »
Woman-Owned: No
Minority-Owned: No
HUBZone-Owned: No
Phase 1
Fiscal Year: 2004
Title: Static Analysis of AI Systems
Agency / Branch: DOD / DARPA
Contract: W31P4Q-04-C-R141
Award Amount: $99,000.00


Toolkits for Artificial Intelligence (AI) are increasingly being used in government and industry. If such systems have access to sensitive information, it is important to know their security properties. A toolkit may contain low-level flaws, such as buffer-overrun errors that allow an attacker to gain control of the host system. Or, there may be flaws in the rule base of a system implemented with the toolkit that allow unauthorized access to sensitive information. We propose the detailed study of the security properties of CLIPS, a widely used expert-system shell in two parts. The first part will be a detailed analysis of the source code of the system using static analysis tools and other methods to find low-level flaws. The second part will be a study of the CLIPS language and the exploration of static and dynamic approaches to create secure CLIPS programs. A promising approach is to model the CLIPS system as a weighted push down system, and to use model-checking techniques to implement a range of security analyses, including termination analysis and information flow. The results will include a design for a tool to be prototyped in Phase II, and guidelines on how to write a secure CLIPS specification.

Principal Investigator:

Paul Anderson
Senior Software Engineer

Business Contact:

Ray Teitelbaum
Small Business Information at Submission:

317 N. Aurora Street Ithaca, NY 14850

EIN/Tax ID: 161338879
Number of Employees:
Woman-Owned: No
Minority-Owned: No
HUBZone-Owned: No