Background & research
The memory safety problem
Why bounds, lifetimes, and authority matter in native software.
Read noteTechnical notes
Memory safety notes
Educational notes on memory errors, software controls, and hardware-assisted protection.
Educational notes on one part of the trustworthiness stack. These explain mechanisms and their limits; they are separate from AGIACC’s research directions.
Start with the memory safety problem, then compare software controls, hardware approaches, MTE, and capability-based protection. Finally, consider the boundaries in AI systems.
Background & research
Software defenses
Languages, sanitizers, fuzzing, and exploit mitigations in a layered design.
Read noteBackground & research
Hardware-assisted protection: different properties
A property-based comparison of memory tagging, control-flow protection, and capabilities.
Read noteBackground & research
Arm MTE in practice
How allocation tags and fault handling affect the memory-protection boundary.
Read noteBackground & research
Capability-based protection with CHERI
Bounds, permissions, and runtime integration as building blocks for isolation.
Read noteBackground & research
Memory safety in AI systems
Native dependencies and component boundaries across AI workloads.
Read note