Submit finding

All systems

Zellic AI / V12 Platform

V12

Zellic's agentic security platform, with accepted findings across the Linux kernel, PostgreSQL, Redis, Dolphin, and Web3 infrastructure.

7
Indexed entries
2
CVE IDs tracked
5
Critical/high entries
91%
Evidence index

What it is

V12 is Zellic’s agentic security platform. Zellic introduced V12 in September 2025 as an autonomous auditor combining AI, LLMs, and traditional static analysis. The current V12 homepage describes the product as agentic security that finds, rates, and reports vulnerabilities.

The initial public positioning focused on Solidity and Web3 audits, but the public record now includes Linux and PostgreSQL CVEs plus accepted no-CVE reports in Redis, Dolphin, wallet, intent, and zero-knowledge infrastructure.

What is verified

The current ledger counts seven V12 entries: Fragnesia (CVE-2026-46300), shared credit on PostgreSQL CVE-2026-14669, plaintext GridPlus pairing credentials in Rabby Wallet, a NEAR Intents storage-exhaustion and garbage-collection bypass, a twelve-report Miden Node campaign, a Redis use-after-free exploited to remote code execution, and a two-bug Dolphin guest-to-host escape chain.

For Fragnesia, independent corroboration comes from Linux and distribution-side records: Ubuntu tracks the CVE as a high-priority local privilege escalation, Debian’s tracker links the V12 PoC and the upstream patch, AlmaLinux released patched kernels, and the netdev patch from William Bowling identifies the skbuff shared-fragment marker root cause.

Why it matters

V12 adds another system to the public AI-assisted vulnerability-discovery record, and the target is not a toy project. A Linux kernel local-root finding in the Dirty Pipe / CopyFail / Dirty Frag page-cache family is exactly the kind of high-impact issue that changes patching pressure for distributors and operators.

The newer no-CVE entries are included because their public reports document maintainer acknowledgment, merged fixes, security releases, or a bug bounty. The AI-attribution detail still comes from V12, so its credits retain the self-reported label even where upstream sources independently confirm the bugs.

Sources

Attributed findings

Catalogued entries credited to V12.

ID Title System Evidence date Severity
v12-redis-sibling-eviction-uaf V12 turns a Redis sibling-eviction use-after-free into remote code execution Heap use-after-free leading to remote code execution - self-reported V12 (self-reported) Aug 17, 2026 high CVE-2026-14669 Codex Security and V12 share credit on PostgreSQL to_char RCE Heap buffer overflow enabling database-server code execution - self-reported OpenAI Aardvark / Codex Security (direct) + V12 (self-reported) Aug 13, 2026 high v12-dolphin-guest-to-host-rce-chain V12 chains two Dolphin DSP-HLE memory flaws into a guest-to-host escape Out-of-bounds stack read and indexed stack write chained into guest-to-host code execution - self-reported V12 (self-reported) Aug 11, 2026 high v12-miden-node-twelve-finding-cluster V12 publishes twelve accepted Miden Node findings Validation, denial-of-service, state-consistency, and protocol-logic flaws - self-reported V12 (self-reported) Jun 30, 2026 high v12-near-intents-storage-exhaustion V12 reports storage-exhaustion and GC bypass in NEAR Intents Storage exhaustion and garbage-collection bypass - self-reported V12 (self-reported) Jun 25, 2026 medium v12-rabby-gridplus-pairing-credentials V12 finds plaintext GridPlus pairing credentials in Rabby Wallet Plaintext storage of hardware-wallet pairing credentials - self-reported V12 (self-reported) Jun 25, 2026 medium CVE-2026-46300 Fragnesia: V12-assisted Linux kernel page-cache LPE CVE-2026-46300 Shared page-fragment marker loss leading to page-cache corruption and local privilege escalation - direct V12 (direct) May 13, 2026 high