Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Strengthen, Prepare, Detect, React (SPDR) to Mitigate the Insider Threat PI Meeting 21 June 2007 Adventium Labs: Tom Haigh, Charles Payne General Dynamics: Johnathan Gohde SRS Phase II PI Meeting 21 June 2007 1 Outline • Insider threat problem and examples • SPDR approach • Accomplishments in past 6 months – Architecture and Evaluation – Scenario implemented and working with Mission Monitor (Demonstration) – Attack planning, attack recognition and sensor generation (Demonstration) – Prototype DREDs with key features operational • Plans for next 6 months SRS Phase II PI Meeting 21 June 2007 2 MI Problem & Examples • Malicous insider (ARDA Workshop, 2005) – Is in organization: Ames (CIA), Hansenn (FBI), Montes (DIA) – Has approved/necessary access, privilege, or knowledge of organization’s information systems, information services, and missions. – Is motivated to adversely impact missions by compromising confidentiality, integrity, and/or availability • Recent insiders: – Information Week (May 10 2007), Leandro Aragoncillo, career Marine with a top secret security clearance, served under two vice presidents in the White House (2 yrs) and then the FBI (4 yrs), stole classified information in an attempt to foster a political coup in the Philippines, his home country. – The Washington Times (April 6, 2007) Richard Sylvestre, U.S. Navy contractor sabotaged a national security computer network at a Navy command center in Italy. A criminal complaint, “would have shut down the entire network that tracks the locations of ships and submarines.” – CICentre.com (March 7, 2007), Paul R. Hall, signalman on USS Benfold, e-mailed Al Qaeda contact, advising that Battle Group would pass through Straits of Hormuz in 19 days and would be highly vulnerable to small weapons such as RPGs. SRS Phase II PI Meeting 21 June 2007 3 SPDR Solution Workstation with Embedded DRED Security Processor Inspect Proxy Honeypot Filter Encrypt (Malicious) Insider SPDR Reasoning Engine Attribution Response Mission & Provenance Monitoring Attack Plan Recognition Attack Plan & Sensor Generation Attack Plan and Mission Models All Network Access controlled by Detection & Response Embedded Device (DRED). Restricts range of feasible attack plans, hence reducing amount of reasoning required. SRS Phase II PI Meeting 21 June 2007 4 Accomplishments • Requirements, Architecture, Design, & Evaluation – Submitted draft CDRL A003, 13 Feb 07 – Added Mission & Provenance Monitors, Attribution Engine • Improved attribution • Will include in July update of A003 • Initiated collaboration with BBN – Both using Adventium’s network model – Began discussion of integration options • Defined and implemented evaluation scenario & mission monitor – Carrier-base Anti-Surface Warfare (ASuW) mission planning – Based on GD tactical SOA – Demonstration today • Restructured attack plan and sensor generation – Improved efficiency and scalability – Demonstration today – Interface with plan recognition module (YAPPR) • Operating DREDs – Key functions operational – Several form factors SRS Phase II PI Meeting 21 June 2007 5 SPDR Architecture Off-line Policy Database Plan Library Plan Generation Plan Library Sensor Configuration Attribution Mission Progress Plan Recognition Per user Policy Plan Likelihood Implemented Responses Mission Monitor Event Traps Provenance Monitor Mission Progress Response Sensed Events Online Response Commands Distributed Security Modules SRS Phase II PI Meeting 21 June 2007 6 Signed Metadata Sensor Configuration Mission & Provenance Monitors • MM tracks the progress of each mission with respect to its plan – Uses functional dependency and timing model of the mission – Receives mission message notifications from DREDs – Reports mission step anomalies or delays to the PRM ● PM preserves message data & metadata that can be used, if a mission fails, to help identify the cause of the failure and the user responsible – Receives message information from DREDs – Stores info in database for use in attribution analysis SRS Phase II PI Meeting 21 June 2007 7 Evaluation Scenario • • • • Anti-surface warfare using Tactical SOA (TSOA) developed by General Dynamics. Carrier-based scenario: threat assessment, flight plan development, hand-off to E-2C. Embedded in general network traffic to extend attack surface 2 classes of insider: – On carrier network but outside mission – Inside mission SRS Phase II PI Meeting 21 June 2007 8 Evaluation Testbed – Summer 2007 Air Ops Strike Ops CDC Intel Ops Workstation Workstation Workstation Workstation (Dell) (Dell) (Dell) (Dell) DRED-1 DRED-2 DRED-3 DRED-4 (Sunfire) (Sunfire) (VIA-C3) (VIA-C7) card reader FW/Router Smart Switch DRED-x DRED-y (t.b.d.) (t.b.d.) Non-Mission Common Server Workstation EMail, File (Dell) (HP/Compaq) Rogue Workstation (Dell) SPDR Security (Sunfire) SRS Phase II PI Meeting 21 June 2007 9 External Networks 3-Layered Plan Generation Test network model with over 20K instances, built using automated scanner NetBase Ontology (Class Definitions) Build once, use often, for many networks Is being used by BBN for CSISM A X P1 (strategic) Zone level abstract path analysis (100 objects) D Y B C P2 (tactical) Only need detail for 2 zones (1000’s of objects) P3 (observable) Observable host-to-host and intra-host actions (very simple plans) B HB Reduces problem size two orders of magnitude. C Sensor/effector relevant annotations AnnotationA 0xFFBA HC SRS Phase II PI Meeting 21 June 2007 10 Generates all interesting plans in a few seconds. Example Level 1 & 2 Outputs Top Level Goal g1: steal ssh_priv_key_edc… Level 1, Plan 13 : [e,k,g,h] e: vint(zone_client, zone_wireless, c_wap_service(zone_wireless), credz__J5) g: read(credz_4l5, zone_internal, c_ssh_service(zone_internal), sshd_rexec_exploit) h: read(ssh_privkey_edcfc828b41a41fe12cf476b721bc279ec657b31, zone_internal, c_ssh_service(zone_internal), credz_4l5) k: vint(zone_internal, zone_client, c_smtp_service(zone_internal), smtp_redirect_exploit) 92 Level 2 expansions: ( ( S1 S2 ) ( S3 S4 ) ( S5 S6 S7 ) ( S8 S9 S10 S11 S12 )) ( ( S1 S2 ) ( S3 S4 ) ( S5 S6 S7 ) ( S8 S9 S10 S13 S14 S12 )) ( ( S1 S2 ) ( S3 S4 ) ( S5 S6 S7 ) ( S8 S9 S10 S11 S12 )) ( ( S1 S2 ) ( S3 S4 ) ( S5 S6 S7 ) ( S8 S9 S10 S13 S14 S12 )) ( ( S1 S2 ) ( S3 S4 ) ( S5 S6 S7 ) ( S15 S16 S17 S18 S19 S12 )) ( ( S1 S2 ) ( S3 S4 ) ( S5 S6 S7 ) ( S15 S16 S17 S18 S19 S12 )) ( ( S1 S2 ) ( S3 S4 ) ( S5 S6 S7 ) ( S8 S9 S10 S11 S12 )) … S1 : connect(iface(med_gravity_wlan, ip_192_168_13_13), gravity_wap_service) S2 : create_if(con(iface(med_gravity_wlan, ip_192_168_13_13), gravity_wap_service), labs_wep_secret, med_labs_switch2, ip_10_0_1_2) … Plan 8. : 6 expansions Plan 5. : 92 expansions SRS Phase II PI Meeting 21 June 2007 11 Plan 1. : 6 expansions YAPPR Plan Recognition Module • YAPPR – Yet Another Probabilistic Plan Recognizer – – – – SIFT, U. Edinburgh, DARPA Integrated Learning Input: library of hierarchical task network plans (AND/OR graph) Compiles plan library into form suitable for rapid online recognition Online: sequentially estimates probabilities of possible intentions given observation stream • Tests on YAPPR scalability – Randomly generated plan libraries of various sizes – e.g. “Large” plan lib (near limits of current YAPPR implementation): • 10 high level intentions (HLI) • 10 distinct high level plans (HLP) of 10 steps each per HLI • 10 distinct low level plans (LLPs) of 10 steps each per HLP – Compilation times (an offline activity) on the order of 1 minute – This is larger than we expect real plan library to be for AsuW scenario • Planner ---> YAPPR interchange format – Plan library is emitted by planner in this form. – Translator to YAPPR HTN representation is being tested at SIFT SRS Phase II PI Meeting 21 June 2007 12 DRED in a Nutshell SRS Phase II PI Meeting 21 June 2007 13 DRED: Prevent • Focus is on enforcing authorized behavior to thwart insider attack • Authorized behavior is what is required for this user for this mission • Prevention opportunities exist for network through application layers SRS Phase II PI Meeting 21 June 2007 14 DRED: Detect • Focus is on discovering events that may suggest an insider attack and on providing sufficient evidence to support (later) attribution of an attack. • DRED manager collects all events (except Mission Status) • Some event consolidation by the DRED is anticipated for efficiency SRS Phase II PI Meeting 21 June 2007 15 DRED: Respond • Focus is on thwarting misuse and on supporting attribution • Each response is determined by the Response Module but coordinated through, and directed by, the DRED Manager • Response may also include additional logging without further restriction SRS Phase II PI Meeting 21 June 2007 16 DRED: Manage User-based Policy (dynamic) • DRED-based Policy (static) Focus on what can be managed easily – pf makes an effective 'policy router' • Split policy into static and dynamic parts (see above) • Leverage Adventium's policy management strategy for distributed policy enforcement points [DARPA OASIS Dem/Val, DHS EFW/VPG] SRS Phase II PI Meeting 21 June 2007 17 DRED Status • • July – diverse implementations January – log collection and consolidation • Sunfire Bump-in-the-wire (2) • Mungo-based breadboard (1) – integration with DSM Manager and Mission Monitor – performance tests – user-based pf rules – IPsec between multiple DREDS using X.509 authentication – simple HTTP proxy – snort, honeyd • September – proxy configuration for message compliance and mission status – DRED policy management – user authentication for userbased policy SRS Phase II PI Meeting 21 June 2007 18 SPDR Plans and Timeline FY-07 Dec Mar FY-08 Jun Sep Dec Mar Jun 1 1: Requirements, Architecture, Design, Evaluation Plan (CDRL A003) 2 3 5 2: Spiral 1 (Component Development) 8 9 3: Spiral 2 (Integrated System) 10 4: Spiral 3 (System Hardening) 4 6 7 11 5: Test and Evaluation 1 2 3 4 6: Program Management & Travel Program Review 1. CDRL A003 Delivered (Feb.) Revisions in July 07, January 08, and June 08. 2. Initial DRED: User-based IP filtering and sensing (Mar.) 3. Initial planner: Strategic and tactical levels (Apr.) NetBase ontology (network model) shared with BBN 4. Scenario defined: Carrier flight planning (Mar.) 6. Functional Model of Scenario (Jun.) Deliverable (update) Final Report 5. Components operational (Jul.) Demonstrations in September 7. Testbed complete (Oct.) 8. Initial integration (Oct.) 9. E2E demonstration (Jan.) 10. Final software build complete (Mar.) 11. Red Team evaluation complete (May) SRS Phase II PI Meeting 21 June 2007 19 Strengthen, Prepare, Detect, React (SPDR) Objective: Thwart and attribute insider attacks by combining AI-based plan generation, sensor synthesis and plan recognition with HW-based, tamper resistant, end-point sensing & control Benefits: • • • Increased accountability Rapid, targeted response to malicious insiders Minimizes impact on benign activities, thereby improving mission effectiveness Research Challenges: Approach: • • • Adapt existing Adventium & SIFT reasoning technologies • Exploit evolving COTS hardware capabilities • Military relevance via GD-AIS application scenarios • • • Thwart plausible, but unknown attacks Identify malicious insiders even when all their actions are authorized Minimize impact on benign activities Ensure sufficient strength of sensing and control mechanism, so insider with physical access cannot disable or avoid it. Maintain compatibility with evolving DoD systems – – – Push security toward the endpoints DoD Enterprise Sensor Grid Strategy GIG vision of end-to-end encryption Innovations: • A priori reasoning anticipates attacks, informs defenses, and generates sensors • Intelligent attack recognition reduces false alarms and supports targeted response creation • Unbypassable, tamper-resistant network endpoint for sensing and control (important for insiders) SRS Phase II PI Meeting 21 June 2007 20 Questions? Comments? SRS Phase II PI Meeting 21 June 2007 21