https://www.usenix.org/conference/usenixsecurity25/presentation/ruegge
The main point is that Intel’s modern Spectre v2 defenses, especially eIBRS, can fail because branch predictor updates happen asynchronously. The researchers show that this timing creates “Branch Predictor Race Conditions” (BPRC), where branch predictions can be learned or applied with the wrong privilege context. In practice, that breaks intended isolation boundaries such as user-to-kernel, guest-to-hypervisor, and even barriers meant to flush unsafe predictions.
The talk’s key attack is called Branch Privilege Injection (BPI). In plain English: an unprivileged process can trick the CPU into treating attacker-controlled branch predictions as if they belonged to a more privileged context, then use a Spectre-v2-style side channel to read privileged memory. The researchers say they built an end-to-end exploit that leaks arbitrary kernel memory from up-to-date Linux systems across six Intel CPU generations.
The practical results are the scary part. In their evaluation, they report arbitrary kernel-memory leakage at 5.6 KiB/s with 99.8% byte accuracy on Intel Raptor Cove, fast KASLR derandomization, and an end-to-end demo that leaked the root password hash from /etc/shadow with a median runtime of 21 seconds. The repository also notes there was a dedicated Black Hat demo for this work.
So the big takeaway of the video is not just “here’s a new exploit,” but “hardware security boundaries are fragile when asynchronous microarchitectural state gets integrated incorrectly.” The talk argues that even protections widely treated as hardware-backed and robust can be undermined by race conditions inside the predictor machinery itself.
On mitigations, the researchers discuss software and hardware responses: replacing exploitable indirect branches with Retpoline-style defenses, disabling alternate return prediction where relevant, or disabling indirect branch prediction in supervisor mode on supported CPUs. They also evaluate Intel-provided microcode updates, and note that in their tested AMD and ARM processors they saw no indication of the same BPRC issue.
A one-sentence summary: the video explains how a race condition in Intel branch prediction can let user code “smuggle” privileged branch predictions across protections like eIBRS, enabling real leakage of kernel memory on fully patched modern systems.