MIT CSAIL researchers have introduced Interrupt Injection, a technique that exploits a tiny timing window between branch predictor sanitization and its actual use. Instead of breaking Spectre v2 directly, it races the mitigation itself by delivering a precisely timed hardware interrupt. The work introduces a broader concept called TONTOU (Time-of-Neutralization to Time-of-Use), suggesting that speculative execution mitigations should also account for asynchronous events like interrupts, not just predictor state. AMD has acknowledged the issue and Linux has already merged a kernel fix to harden SafeRET against this attack. Intel’s assessment differs, stating that existing guidance remains sufficient.