Tiny Chips, Big Leaks: Breaking TrustZone-M with Single-Stepping Attacks

No ratings

Presented at Black Hat USA 2026 by

Trusted execution environments (TEEs) provide confidential-computing guarantees by running sensitive code inside hardware-enforced enclaves that remain isolated even when the operating system is compromised. However, despite this strong architectural isolation, TEEs remain vulnerable to software-based microarchitectural side-channel attacks.In prior work, the researchers have shown that a privileged attacker controlling the untrusted OS can obtain extremely high-resolution, instruction-level observations by precisely triggering timer interrupts after each instruction executed within the TEE. This technique, known as single-stepping, has enabled a wide range of high-resolution side-channel attacks across major TEE ecosystems, supported by dedicated single-stepping frameworks such as SEV-Step (AMD SEV), SGX-Step (Intel SGX), TDX-Step/TDX-Down/TDXploit (Intel TDX), and Load-Step/CacheGrab (Arm TrustZone-A). In contrast, TrustZone-M, the TEE variant designed for low-power, resource-constrained microcontrollers (MCUs), has remained largely unexplored, and prior work has even suggested that Cortex-M platforms are "immune" to interrupt-latency attacks.In this Briefing, we will present M-Step, the first single-stepping framework for side-channel analysis on Arm TrustZone-M. We profile and systematize previously undocumented Cortex-M interrupt-handling behavior at the instruction level, showing how it can be leveraged to reliably single-step production code without debug features enabled. Using M-Step, we uncover new leakage in the latest release of Mbed TLS and demonstrate the first software-based, single-trace attack (CVE-2025-54764) that extracts real-world cryptographic keys from TrustZone-M. Our results show that interrupt-driven microarchitectural leakage is a practical threat on modern MCUs and motivate the need for stronger side-channel defenses in MCU-based embedded TEEs.