Stealthium reports ForgePtr NVIDIA driver root bug, CVE-2026-47501
Stealthium reports ForgePtr (CVE-2026-47501), an NVIDIA driver local root bug; it says driver 615.71.09 fixes it, so update affected hosts.
- Published
- Fixed in
- NVIDIA GPU driver 615.71.09 or later (per Stealthium)
- Exploitation
- No confirmed in-the-wild exploitation; Stealthium says it built a private working exploit and published no proof-of-concept
Live scores
CVE-2026-47501- EPSS
- 0.001
low predicted chance of attack - VEDAS
- 0.053
little exploit evidence - On CISA KEV
- Not listed
- Detection rule
- Out of scope (local)
Thin horizontal line: the “likely exploited” threshold (EPSS 0.10, VEDAS 0.5). Shaded “no data” areas are periods when cve-scores did not publish.
What happened
According to Stealthium, its researchers found a local privilege escalation bug in the NVIDIA driver, which it names ForgePtr and tracks as CVE-2026-47501. Stealthium says it built a working exploit to confirm root and reported the bug to NVIDIA. It says it is not publishing a proof-of-concept and is holding back some specifics. Stealthium
Stealthium describes the root cause as a missing error check in the driver’s event-buffer setup path. It says a caller-supplied handle is converted into a kernel pointer without checking whether the conversion succeeded. When the object is later freed, the kernel decrements a counter at the attacker-supplied address. Stealthium says this can be turned into writing chosen bytes to a chosen kernel address. Stealthium
Stealthium claims that, from an unprivileged local account, this lets an attacker:
- reach root by rewriting the kernel’s
core_patternsetting; - escape an unprivileged container to the host, because
core_patternis global kernel state that is not namespaced; - switch off protections such as ASLR, kernel lockdown and module signature enforcement, and load an unsigned kernel module on a Secure Boot machine.
Stealthium says it verified these steps on affected hardware. It also says IOMMU and GPU confidential-compute features do not apply, because the kernel itself performs the write. Stealthium
On severity: Stealthium itself calls the bug “beyond critical” for organizations renting or running multi-tenant GPUs, and describes full host takeover in that setting. The “medium” rating on this page is a conservative site rating, used because no official source (such as NVIDIA’s bulletin or an NVD score) was available to review. It is not a statement that the risk is small for multi-tenant hosts. Treat those hosts as a priority.
What is not confirmed: the findings come from a single vendor blog post. NVIDIA’s own security bulletin, which Stealthium references, was not reviewed, and there is no independent verification of the exploit claims. No exploitation by attackers in the wild is reported in the source. Stealthium also says it has 25+ further High or Critical vulnerabilities under responsible disclosure, but gives no details.
Am I affected?
According to Stealthium:
- Affected range: every open-source NVIDIA driver release from 515.43.04 (2022-05-09) through 610.57.04 (2026-08-03). It says it confirmed root at runtime on 610.43.02 and 610.57.04.
- Fixed in: 615.71.09 (2026-09-09).
- Requirements: a driver in the affected range, an unprivileged local account (no special group, no CAP_SYS_ADMIN), and an additional unprivileged KASLR leak primitive to resolve live kernel addresses. It says the write primitive is architecture-agnostic.
- Closed driver: Stealthium says the bug likely predates open-sourcing in the closed driver too. This is its inference and is not confirmed.
Stealthium says the risk is highest for hosts that hand GPUs to containers or tenants you do not fully trust. Stealthium
What to do now
- Check the NVIDIA driver version on every GPU host, starting with nodes that give GPUs to containers or other tenants.
- Update to driver 615.71.09 or later. Stealthium says NVIDIA released the fix on September 9, 2026 and published its security bulletin on September 30, 2026. Confirm the details against NVIDIA’s bulletin, which was not reviewed here. Stealthium
- Do not rely on a workaround. Stealthium says none exists short of patching, because the flaw is in the driver’s own logic and no sysctl or seccomp profile closes it. This is Stealthium’s claim.
- For multi-tenant GPU hosts, Stealthium suggests treating the update as urgent.
- Stealthium says exploitation looks like ordinary driver use plus a process crash. Its suggested hunting signals are malformed event-buffer setup calls, unprivileged writes to
core_pattern,kernel_locked_downorsig_enforce, and a container modifying host-wide kernel globals. Whether your tooling can see these is not confirmed. Stealthium