[DigitalToday reporter Chi-gyu Hwang (황치규)] As leading security companies publish their experiences of using Anthropic’s AI model Mythos, which was first released on a limited basis to some companies, Cloudflare’s case stands out.

Cloudflare rated Mythos as "a tool on a different level, not a simple improvement" compared with previous general-purpose frontier models. It singled out two capabilities: building attack chains and generating proofs. Cloudflare also pointed out that Mythos sometimes refused requests for defensive vulnerability analysis. It added that even with the same task, wording it differently sometimes produced the opposite result.

· Cloudflare speaks about the power and limits of Anthropic’s Mythos

In the meantime, Anthropic allows Mythos users to share cyber threat information with other companies exposed to similar vulnerabilities. It is meant to address concerns that restricting access to information could harm small and midsize companies.

· Anthropic allows external sharing of Mythos cyber threat information, revising its existing confidentiality policy

Hacking that exploits AI appears to keep increasing. DigitalAI, a DevOps company, said 87 percent of monitored apps were attacked in 2026. That is a sharp rise from 55 percent in 2022. DigitalAI said this coincides with the spread of AI model use since ChatGPT’s launch in November 2022. A survey also found that exploiting software vulnerabilities as a cause of data leaks overtook credential theft for the first time, with hackers using AI.

· "87 percent of mobile apps were attacked"... AI lowers barriers to entry for hackers, driving a surge in attacks · Data leaks from exploiting software vulnerabilities surge, overtaking credential theft

The security industry is stepping up efforts to strengthen AI-based defenses to counter hacking threats that misuse AI. Microsoft introduced an agentic security analysis system called the Multi-Model Agentic Scanning Harness (MDASH), covering vulnerability discovery, validation, proof and response. Microsoft said AI vulnerability discovery is moving beyond the research stage into an engineering task, and predicted results could scale. Cybersecurity startup Depthfirst said its AI model found many bugs that Anthropic’s Mythos missed. Depthfirst claimed the cost was about one-tenth of Mythos.

· Microsoft: "AI vulnerability discovery is moving beyond the research stage"... scaling signaled · Depthfirst: "We found more bugs Mythos missed, and at one-tenth the cost"

The following is a roundup of moves and issues around security involving companies in and outside South Korea.

GitHub, a well-known developer platform under Microsoft, suffered an incident in which data was stolen from about 3,800 internal code repositories in a hacking attack. The breach started on an employee device with a malicious Visual Studio Code extension installed.

· Data stolen from 3,800 GitHub internal repositories... malicious extension on employee device blamed

Autonomous security operations company Toruq acquired cybersecurity startup Jit.io. Founded in 2021, Jit has technology that creates an AI-based context graph for use in security investigations. Toruq plans to integrate Jit’s technology into its AI SOC (Security Operations Center) platform. Cloud computing and security company Akamai agreed to acquire LayerX, a company specialising in AI and browser security, for about $205 million.

· Toruq acquires Jit, expanding its organisation-context security operations platform · Akamai to acquire AI and browser security company LayerX for $205 million

Logpresso officially launched its AI-based XDR platform, Logpresso Sonar 5.0. Logpresso is expanding AI beyond a simple assistant tool into a core technology spanning threat detection, analysis and response automation, and is accelerating its push into the market for autonomous integrated security operations platforms. ITCEN PNS signed a partnership agreement with global security software company BlackBerry and will supply BlackBerry’s main security and crisis response solutions in the South Korean market. Igloo Corporation secured two new technology patents related to implementing an Autonomous Security Operations Center, or Autonomous SOC. Jobis&Villains, which operates tax refund platform SamJjeomSam, is introducing a global integrated security system and upgrading its threat detection framework. HancomWith is joining the "2026 zero trust adoption pilot project" jointly promoted by the Ministry of Science and ICT and the Korea Internet & Security Agency (KISA) through a consortium with Amjin as lead company, SK Broadband, Basestone and DST International.

· Logpresso launches XDR platform Logpresso Sonar 5.0, also partnering with StealthMole · ITCEN PNS to supply BlackBerry security and crisis response solutions in South Korea · Igloo Corporation secures 2 AI and SOAR patents for implementing an autonomous SOC · SamJjeomSam introduces an integrated security monitoring system, expanding prevention-focused security · HancomWith joins zero trust pilot project, supporting identifiers, identity and data domains

KT set up a Personal Information Protection Advisory Committee to respond to changes in the personal data processing environment, restore customer trust and build an ecosystem for responsible data use. KT’s advisory committee is strategic governance aimed at strengthening a prevention-focused personal data management system. KT is participating in the 2026 pilot transition support project for quantum-resistant cryptography and will apply post-quantum cryptography (PQC) to key defence systems. Western Digital integrated post-quantum cryptography, or PQC, into its latest high-capacity Ultrastar UltraSMR hard disk drives.

· KT establishes a Personal Information Protection Advisory Committee, shifting to prevention-focused measures · KT to apply quantum-resistant cryptography security to defence systems · Western Digital to integrate quantum-resistant cryptography technology into hard disk drives