Interpreter Memory Safety via Differential Fuzzing with a CHERI on Top
summary
A differential fuzzing framework called diffCHERI:FruitFly generates and mutates MicroPython programs with an LLM and CST operators, then runs them on conventional x86/ARM builds and a CHERI Arm Morello port. Over 24 hours it executes 8,189 programs, uses delta debugging to minimize cases, and reports 40 defects on conventional builds and 51 in the CHERI port. CHERI turns silent corruption into CHERI capability faults, giving precise, hardware-enforced localization that improves bug-finding.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:05Research paper overview
- 3:13High-level on microcontrollers
- 4:20CHERI fundamentals
- 6:30Differential fuzzing
- 7:56LLM buggy Python
- 9:41Dual-lane bug detection
- 12:32Evaluation setup
- 13:54Bug examples impact
- 17:26CHERI debugging role
- 18:11Limitations and threats
- 19:36Conclusion future work
- 20:19Q&A
- 20:40Formal verification question
- 21:42LLM bugs question
- 22:43Closing
related talks
31mMemory Safety: The Pernicious Challenge
01h 33mHands-on Introduction to CHERI
10mCHERI in out-of-order microarchitectures
20mMaking CHERI Accessible
11mFast and fine-grained compartmentalisation in CHERI
25mLightning Talk: Protecting Zephyr Against Memory Safety Vulnerabilities With the New CHERI Hardware Architecture- 25mLightning Talk: Protecting Zephyr Against Memory Safety Vulnerabilities With the New CHERI Hardware Architecture
18mWhat’s next for memory safety?