AI Breaks Every Cyberattack Benchmark — Security Researchers Warn

According to CyberScoop, researchers who evaluated the autonomous cyber capabilities of the latest AI models (GPT-5, Claude, Mythos) confirmed that they **broke every existing benchmark**. The fact that AI's capacity for autonomous cyberattack has now surpassed conventional evaluation criteria is raising serious concerns across the security field.

This suggests that AI is transitioning from a mere tool to a stage where it can autonomously discover and exploit vulnerabilities, and defensive capabilities now demand a fundamental rethink.

Party Leader Anno Criticizes Slow Government Response to Mythos — "I Thought We Shared a Sense of Crisis"

At a press conference on May 14, Takahiro Anno, leader of Team Mirai, pointed to the Japanese government's slow response to "Claude Mythos Preview", an AI model from the US company Anthropic. Anno said he had until now shared with the government his sense of crisis over the rapid progress of AI, and expressed concern that an official response to the arrival of Mythos has been slow.

His remarks once again called into question the state of AI literacy and crisis management in Japanese politics.

US and China in Talks to Prevent AI Proliferation to Non-State Actors

According to CNBC, it has come to light that the United States and China are holding talks to prevent AI models from falling into the hands of non-state actors such as terrorist organizations and criminal groups. Led by the US Treasury Secretary, the dialogue signals a willingness on both sides to coordinate on managing the security risks of AI.

Regulating the military and security use of AI has become an international challenge, and the outcome of these talks could influence AI regulation in other countries going forward.

Meta Rolls Out AI "Secret Mode" on WhatsApp

In an official blog post, Meta announced that it has introduced "Secret Mode (Incognito Mode)" to Meta AI chats within WhatsApp. The feature lets users keep their conversations with the AI in an encrypted state, so they can use the AI assistant while still protecting their privacy.

Privacy protection in AI services is one of consumers' top concerns, and it is noteworthy that a major platform has built end-to-end encryption in as a standard feature.

"Body Theft" in Deepfake Porn — Victims Speak Out

MIT Technology Review has published a detailed report on a new dimension of the harm caused by deepfake pornography. While past discussion has focused mainly on victims whose "faces" are used without consent, there are also victims whose "bodies" are used without permission. One woman discovered a deepfake in which someone else's face had been composited onto her body from an old video. "It feels like wearing someone else's face as a mask," she said.

The problem of non-consensual intimate images (NCII) is growing more severe as technology advances, making legislation and platform action an urgent priority.

65% of Girls See AI as a "Friend" — Girl Scouts Survey

A survey conducted by the Girl Scouts found that 65% of girls who use AI-equipped devices view AI as a "friend". Debate is intensifying among educators as well over how AI affects children's social relationships. Long-term research is needed on how building pseudo-relationships with AI affects child development.

South Korean Markets Rattled by Remarks on an AI-Driven "Citizen Dividend"

Yahoo Finance reports that remarks posted on Facebook about a "Citizen Dividend" scheme to distribute AI-generated profits to citizens threw South Korean markets into confusion. The episode is drawing attention as a case in which policy debate over AI-driven economic transformation has moved actual markets.

---

*This article was compiled from collected news data. Please refer to the original articles for details on each topic.*