[DigitalToday reporter Chi-gyu Hwang] As AI evolves, concerns are growing that cyber attackers could abuse it. Among AI model developers, there appears to be a growing move to provide new models on a limited, early basis to selected companies, including security firms, so they can prepare for cyber attacks.

Anthropic introduced Project Glasswing to support selected companies in using its AI model under development, Claude Mythos Preview, first for security purposes. Anthropic says competitors can also join Glasswing. It also left open the possibility of OpenAI joining.

Anthropic releases Claude Mythos Preview first to some companies to check risks of misuse by hackers. Anthropic says competitors, including OpenAI, can participate in the Glasswing program.

Anthropic is warning that Claude Mythos and related models could find security vulnerabilities and mount attacks at speeds that cannot be compared with existing models.

"AI agent hackers are coming... a turning point in cybersecurity history."

Anthropic's particular focus on security is also linked to Mythos' advanced reasoning capabilities. Platformer founder and editor Casey Newton (케이시 뉴턴) said, "Existing models were at the level of finding vulnerabilities, but Mythos can find five vulnerabilities in a single piece of software and link them into a chain attack." He also conveyed Anthropic's view that combined with the ability for AI to keep working alone for long periods without humans, it could become an inflection point for cybersecurity threats.

Why Anthropic says Claude Mythos is an inflection point for cybersecurity threats.

"Existing models were at the level of finding vulnerabilities, but Mythos can find five vulnerabilities in a single piece of software and link them into a chain attack," he said. He also conveyed Anthropic's view that combined with the ability for AI to keep working alone for long periods without humans, it could become an inflection point for cybersecurity threats.

OpenAI also plans to provide a new model with advanced cybersecurity capabilities on a restricted, early basis to only a small number of companies. In February, after the launch of GPT-5.3-Codex, OpenAI introduced an invitation-based pilot program called Trusted Access for Cyber. At the time, OpenAI said it would provide participating companies with model access and $10 million in API credits so they could use it for defensive security work.

OpenAI, like Anthropic, to release new model on a limited basis over security threat concerns.

As AI-based cyber attacks surge, calls are also growing to fundamentally change how corporate leaders respond. IBM research showed cyber attacks targeting externally accessible software and system applications rose 44 percent in a year, and many were linked to AI-based vulnerability exploitation.

AI-based cyber attacks jump 44 percent in a year; 77 percent of companies lack even basic security systems.

The article also summarizes moves by companies in South Korea and abroad related to AI.

At the 2026 Japan IT Week Spring in Tokyo, Hancom signed MOUs with CyberLinks and Sun Digital System to jointly push biometric authentication and authentication system businesses targeting Japan's finance and enterprise markets. It is also pursuing cooperation with Money Partners Solutions on overall digital trust solution member management and membership screening systems. Genian is rapidly expanding its customer base, mainly in overseas markets, and has surpassed 200 global cumulative customers. Naru Security signed an MOU with hosting services company Viaweb to develop and cooperate on a ZeroTrust Hosting model for small and midsized companies.

Hancom accelerates push into Japan with AI and biometric authentication. Genian surpasses 200 global customers. Naru Security and Viaweb launch ZeroTrust Hosting for small businesses.

TeamStone, a company specializing in integrated IT control solutions, signed a domestic partnership with DeepKeep, an end-to-end AI security platform company. The two companies plan to respond to AI security issues arising in corporate environments where adoption of generative AI and large language models is expanding, and to step up efforts to target the domestic market.

TeamStone to supply DeepKeep solutions in South Korea.

Check Point Software Technologies is introducing Check Point AI Defense Plane, an integrated AI security control plane designed to let companies manage how they connect, deploy and operate AI across their businesses.

Check Point launches AI Defense Plane, offering end-to-end security targeting AI agents.

Meta halted all ongoing projects with Mercor, a provider of AI training datasets that suffered a security incident, and began investigating the incident. The incident at Mercor appears to be linked to a supply chain compromise involving a LiteLLM update. LiteLLM is a standard adapter library used to call major AI services such as OpenAI, Anthropic and Google, and an attacker known as TeamPCP is reported to have distributed a tainted update.

AI training data supply chain breach: Meta halts all projects with Mercor.

As ransomware attacks increase, more companies are seeking professional negotiators to negotiate directly with hackers. With cyber attacks against major companies worldwide surging, demand is rising for ransomware negotiators held by large security companies such as Palo Alto Networks and Sophos, the report showed.

As ransomware breaches spread, the world of ransom negotiators draws attention.