Why Should Critical Systems Be "Divided"? What IT Departments Should Prepare for in an Era When AI Attacks Can't Be Fully Defended Against

What happened

An article in Japan's IT press argues that AI is now very good at finding security weaknesses. An attacker can use AI to find a hole and attack a system in a very short time. The article says companies should plan for the day an attack gets through, and limit how much damage it can do. One key idea is to split big systems into smaller parts that are kept separate from each other.

Key numbers and facts

• In a test reported on April 27, 2026, an AI startup said that the top AI model from OpenAI got a score in the top 0.2 percent for the Japanese university entrance exam. (The score is from the article's source, not a claim we checked.)
• Attacks that used Anthropic's Claude Code AI tool were reported to do about 80 to 90 percent of the work by themselves, including finding a weakness and getting a password or login.
• Before a company notices an attack, the time it takes to find and use a weakness may now be up to 24 hours shorter than before.
• In some reported cases, the time from first contact to the start of an attack was as short as 2 hours.

Why this topic is new

Many companies buy computer systems from IT firms called system integrators (SIers). These firms build and run systems for their clients. The article says that because of tight budgets, labor shortages, and fast digital change, clients and SIers now often have a weak working relationship. Many systems have grown big and tangled, which makes them hard to protect.

Why it matters

If AI makes attacks faster and cheaper, a single big system is a big target. If one part breaks, the whole company may stop. Splitting systems into smaller, separate parts can limit the damage. It also gives IT teams time to notice a problem and respond. The article suggests that managing a list of software parts (called an SBOM) is a useful first step, so teams know which parts have known weaknesses.

What to watch

• Whether companies begin to split their core systems into separate parts.
• How quickly software makers fix known weaknesses and send updates.
• How AI tools are controlled, and whether powerful AI models for security are released to the public.
• New reports on attacks that use AI agents to run on their own.