Ledger Hardware Wallet Supply Chain Compromise Results in $93.4M Crypto Drain
In mid‑2025, Ledger Nano X/S devices sold through authorized reseller CryptoBilis were found to contain a covert hardware implant—a soldered SMD chip or modified PCB component—that exfiltrated seed phrases and enabled unauthorized transaction signing when connected to a host. Blockchain forensics traced over $93.4 M in drained assets across hundreds of addresses, linking the thefts to a supply‑chain tampering campaign that extended from Southeast Asia to Europe. Ledger halted CryptoBilis sales, issued user warnings, and launched a forensic audit.
NVIDIA Open Agent Safety Platform OASP HardwareBased Agent Governance
NVIDIA unveiled the Open Agent Safety Platform (OASP) in September 2026, coupling the open‑source OpenShell runtime with the Sentry watchdog reference design that runs on BlueField‑4 DPUs. OpenShell provides kernel‑level isolation, sandboxed execution, and per‑outbound‑request policy checks, while Sentry monitors agent behavior out‑of‑band and can quarantine or halt malicious agents within milliseconds. The platform targets governance of agents on enterprise‑controlled infrastructure, aiming to move enforcement outside the model and into hardware. Analysts estimate it addresses less than 25% of enterprise agentic risk, leaving SaaS, third‑party, and attacker‑introduced agents ungoverned.
GPUThor: Rowhammer Attack Bypasses ECC on NVIDIA RTX A-Series GPUs
University of Toronto researchers have demonstrated GPUThor, a sophisticated Rowhammer-based attack targeting GDDR6 memory architectures in NVIDIA Ampere workstation GPUs, specifically the RTX A4000 through A6000 series. By utilizing non-uniform row hammering patterns, the exploit induces multi-bit flips—specifically double and triple bit errors—that exceed the correction capabilities of standard Error Correction Code (ECC) mechanisms. This bypass allows an attacker to corrupt memory page tables, facilitating a transition from unprivileged program execution to host-level root shell access. The attack demonstrates a massive increase in efficiency, reducing exploit time from nearly 22 hours to approximately 1.1 minutes, posing a significant risk to multi-tenant AI/ML cloud environments and high-performance workstations.