AI News Daily 2026-10-20
Contents
- A GMO Internet Group-affiliated security firm confirmed that intrusions targeting multiple Japanese companies across several industries used an AI agent tool called "ARTEX," reportedly built by a Chinese engineer.
- Chief Cabinet Secretary Kihara publicly addressed the finding at an October 20 press briefing, saying the government will share the information with investigative authorities and work to clarify the background, including whether state involvement is behind the attacks.
- Microsoft announced "Project Perception" and "Guardian Agents," two new capabilities meant to discover, identify and continuously monitor AI agents operating inside a company's own systems, arguing that an agent can become a risk — an "agentic insider" — even without being hijacked by an outside attacker.
01Confirmed use of Chinese-developed AI agent software "ARTEX" in cyberattacks on Japanese firms
Published: 2026-10-20 13:42
Facts
A security company affiliated with GMO Internet Group said it has confirmed that an AI agent program called “ARTEX,” reportedly developed by a Chinese engineer, was used in unauthorized-access incidents targeting multiple companies across different industries in Japan. At a press conference on October 20, Chief Cabinet Secretary Kihara referred to this finding, saying the government would pass the information on to investigative authorities and work to clarify the circumstances, including the possibility that the attacks have a state-linked background.
Background
Bloomberg, reporting on the same finding, described ARTEX as a tool that had previously been used in attacks on South Korean banks before turning up in incidents in Japan, according to GMO. The case adds to a small but growing set of real-world examples in which AI agents — software capable of carrying out multi-step tasks with limited human oversight — are being deployed by attackers rather than by the defenders trying to stop them.
Implications
If intrusion tooling increasingly relies on AI agents to automate work that previously required a skilled human operator, Japanese companies may need to accelerate their own use of AI-assisted monitoring and detection simply to keep pace on defense. The government's decision to address the case publicly, and to pursue the question of a possible state-linked background, also signals that this is being treated as more than a routine cybercrime matter.
Sources: Nikkei — GMO-affiliated firm confirms Chinese-made AI software used in domestic unauthorized-access incidents · Bloomberg — AI Tool That Hacked Korean Banks Was Used in Japan, GMO Says
02Microsoft unveils "Project Perception" and "Guardian Agents" to govern in-house AI agents
Published: 2026-10-19 22:00
Facts
In an official blog post titled “Securing the agentic insider,” Microsoft announced two new security capabilities for organizations running AI agents internally. “Project Perception” is designed to discover agents operating inside a company's systems and assign each one an identity and a defined set of permissions. “Guardian Agents” continuously monitors the risk those agents pose and responds dynamically as conditions change.
Background
Microsoft's framing is notable: the company argues that an AI agent can become a risk — what it calls an “agentic insider” — even without being compromised or hijacked by an outside attacker. In other words, the danger is not only that someone hacks an agent, but that an agent with broad, poorly tracked permissions can become a liability simply by operating inside a company's systems.
Implications
For organizations moving AI agents from pilot projects into full production use, the announcement underscores that conventional, human-centric identity and access management was not built for software that can act autonomously across many systems at once. Governance built specifically for agents — discovery, identity assignment, permissioning and continuous monitoring — looks set to become a baseline requirement for enterprise AI deployment rather than an optional extra.
Source: Microsoft official blog — “Securing the agentic insider”
Editor's note
Today's two stories sit on opposite sides of the same coin. In the ARTEX case, an AI agent is the attacker's tool, automating steps in an intrusion against Japanese companies. In Microsoft's announcement, AI agents are the thing that needs to be governed and watched inside a company's own walls, whether or not an outside party ever touches them. Read together, they suggest that as AI agents become more capable and more widely deployed, the security conversation is shifting from asking whether an agent has been compromised to asking what an agent is capable of doing, and who is watching it.
It is also worth noting what did not make today's report. A number of business and infrastructure stories were circulating during the collection window — including a large funding report for OpenAI, a reported sanctions matter tied to a commercial-secrets case involving Google, a sharp share-price move at a Korean data-center operator, and rising credit-default-swap spreads tied to SoftBank — but none of them could be corroborated by a second independent outlet within the applicable time window, so they were held back rather than published on single-source reporting.