360 Unveils China’s “Mythos” AI Security Agent at ISC 2026

Background

For the past 30 years, the balance of cyber offense and defense has relied on the difficulty of finding vulnerabilities. High‑value bugs are scarce, discovery costs are high, and only a limited number of top security experts and nation‑level teams can sustain systematic exploitation.

Mythos as an AI‑era cyber weapon

Anthropic’s most powerful model, Mythos, can autonomously discover vulnerabilities, analyse them, and even generate network‑attack tools. This capability is described as an “AI‑era cyber nuclear weapon” that creates a new strategic deterrent and delivers a dimensionality‑reduction strike against traditional security measures.

AI’s impact on vulnerability discovery

Artificial intelligence makes vulnerability discovery faster, cheaper, and scalable, turning a scarce expert‑driven activity into a mass‑produced process. Large numbers of long‑standing, hard‑to‑detect bugs can be uncovered systematically, fundamentally rewriting the industry’s rules of the game.

360’s AI security capabilities

At ISC.AI 2026, 360 announced two core AI‑security capabilities under the “Yitian Tulong” brand:

“Tulong Feng” – an automated vulnerability‑discovery agent positioned as a Chinese counterpart to Mythos.

“Yitian Zhen” – an automated defence system for security operations.

Tulong Feng details

Since its launch, Tulong Feng has:

Discovered a total of 3,432 vulnerabilities.

Had 105 of those confirmed by regulators.

Included multiple high‑risk entries in the national vulnerability database.

Covered scenarios such as open‑source code, operating systems, office software, and AI‑agent platforms.

The system transforms the previously expert‑dependent vulnerability‑mining workflow into a sustainable, verifiable, and scalable “factory‑production” process.

Yitian Zhen details

Yitian Zhen operates in real network environments, autonomously planning tasks, analysing alerts, and orchestrating responses. It shifts security operations from manual monitoring, analysis, and response toward automated, intelligent workflows.

Strategic implications

Because vulnerabilities are effectively endless, even a Chinese‑style Mythos cannot eliminate all risk. The proposed defence model is to “counter compute with compute,” moving from a “human‑sea” tactic to an “autonomous‑driving” approach.

Rather than copying foreign models that rely solely on massive compute, the strategy emphasises leveraging engineering advantages to integrate large‑model capabilities, expert knowledge, and vulnerability knowledge bases into a coordinated intelligent‑agent ecosystem.

Pan Stone Shield collaboration

The “Pan Stone Shield” initiative opens the Yitian Tulong capabilities to key Chinese trusted‑innovation enterprises and critical‑infrastructure units, creating a collaborative defence ecosystem. Initial participants include:

Tongxin (统信)

Kirin (麒麟)

Shanshi NetTech (山石网科)

HaiGuang (海光)

FeiTeng (飞腾)

Kingdee (金蝶)

BiRen (壁仞)

Mobile Cloud (移动云)

BaolanDe (宝兰德)

DaMeng (达梦)

The collaboration mirrors the U.S. approach of embedding Mythos in the Glasswing alliance, asserting the need for an indigenous security‑collaboration framework rather than waiting for risks to materialise.