Anthropic to Provide "Mythos" Access to EU Cyber Agency

Anthropic has revealed plans to provide ENISA, the European Union's cybersecurity agency, with access to "Mythos," a powerful AI security tool. Mythos is an AI tool capable of discovering and exploiting vulnerabilities in computer systems, and it is currently being offered to key institutions through an early testing program called "Project Glasswing." With ENISA's participation, the EU's cyber defense capabilities are expected to be significantly strengthened.

This move shows that the serious application of AI to the security field is accelerating. At the same time, how to manage the risk of such tools being repurposed for attacks remains a critical challenge.

3D Model Generation Startup Founded by Gen Z Gamers Becomes a Unicorn in China

In China, a 3D model generation startup founded by Gen Z gamers has grown into the latest AI unicorn. The company, reported by Bloomberg, has rapidly increased its valuation with 3D content generation technology that draws on gaming expertise.

China's AI startup ecosystem is diversifying, driven largely by young entrepreneurs. 3D generative AI is expected to see demand across a wide range of fields, including gaming, architecture, and manufacturing, and its future market expansion is drawing close attention.

NVIDIA Announces New AI Chips for Personal Computers

NVIDIA has announced a new AI chip designed for personal computers. The company plans to fully enter the consumer market with AI inference chips, which have so far been aimed primarily at servers and data centers.

This move signals that the shift toward running AI processing not only in the cloud but also on edge devices is accelerating. Local AI execution offers significant benefits in terms of privacy protection and reduced latency.

Security Boundary Issues Posed by Background AI Agents

A research team at OriginHQ has pointed out that AI agents running in the background are creating new security boundary issues. Agents that continue to execute tasks autonomously may exhibit behavior that falls outside the assumptions of traditional security models.

In particular, when agents operate for extended periods, risks such as unintended privilege escalation and improper access to data have been flagged. As AI agents become more widespread, establishing safe operational frameworks has become an urgent priority.

SAAS: A New Method for Autonomously Suppressing "Over-Search" in Agentic Retrieval

"SAAS (Self-Aware Reinforcement Learning for Over-Search Mitigation)," a study that has attracted attention on Hugging Face's Daily Papers, proposes a new framework in which AI agents recognize the limits of their own knowledge and autonomously suppress unnecessary searches.

Conventional agentic search systems have long suffered from the problem of "over-search," repeatedly calling external retrieval even when it is unnecessary, which increases inference latency and computational costs. SAAS dynamically models search boundaries and introduces a reward design that penalizes unnecessary and redundant searches at the trajectory level. It significantly reduced over-search while maintaining accuracy.

AI Fiction Is the "New Fast Food"

An opinion piece in the Washington Post describes AI-generated fiction as the "new fast food," sparking debate. While AI-generated works that can be mass-produced with ease satisfy consumer demand, there are concerns about the dilution of literary value and the impact on creative culture.

The topic has also generated lively discussion on Hacker News, with a wide range of opinions exchanged, including the view that it represents "the same structural problem as junk food in a mass-consumption society" and counterarguments that "having more choices is itself something to welcome."

---