Researchers at MIT CSAIL have demonstrated a new CPU side-channel attack called TONTOU (Time-of-Neutralization to Time-of-Use) that slips through the Spectre v2 defenses Intel and AMD have spent years building. The attack works by injecting an interrupt at the precise moment between when a CPU clears its branch predictor (the Spectre v2 mitigation) and when the processor actually consults that predictor — a window measured in nanoseconds that is too short to close in software. By poisoning the predictor inside that gap, an attacker with limited local access can steer speculative execution into kernel memory.
On AMD Zen 2 hardware, the researchers achieved a leak rate of 5.47 bytes per second with about 92% accuracy. In a proof-of-concept demonstration presented at Black Hat USA 2026, they extracted password hashes directly from the Linux /etc/shadow file — data that is normally readable only by root. TONTOU bypasses both Intel’s Enhanced Indirect Branch Restricted Speculation (eIBRS) and AMD’s Safe RET mitigation, which are the primary kernel-level defenses against branch-target injection attacks. The research will be published in full at USENIX Security 2026. The researchers reported their findings to AMD and Intel prior to disclosure; no CVE has been assigned yet.
How to check if you’re affected
Affected devices are computers running AMD Zen 2 processors (Ryzen 3000 / EPYC Rome series) on Linux. The researchers primarily tested this hardware, though the underlying timing gap may exist on other microarchitectures. If you run AMD Zen 2 hardware in multi-tenant or shared environments — cloud VMs, academic computing clusters, or servers where unprivileged local users can run code — treat this as high severity until vendor patches or microcode updates are available. Intel hardware is also discussed as potentially affected, though exploitation specifics differ.