---
title: Anthropic co-founder discusses potential cyber attacks involving AI
url: https://www.elseif.net/anthropic-co-founder-discusses-potential-cyber-attacks-involving-ai
published: 2026-10-16T19:04:04+00:00
language: en
section: Agents
source: https://www.itmedia.co.jp/aiplus/article/2610/14/2000002272/
organizations: anthropic
publisher: elseif
---

# Anthropic co-founder discusses potential cyber attacks involving AI

On October 14, a co-founder of Anthropic highlighted concerns regarding potential cyber attacks targeting Japanese companies, suggesting the possibility that AI might be involved. Jack Clark, who also leads the public benefit division, made these comments during a media briefing marking the first anniversary of Anthropic Japan's establishment.

Clark noted that there is a possibility that hacker groups may be exploiting publicly available open-weight models in these recent cyber incidents. He stated, "The details are unclear, but it is likely that AI tools are being used in some form."

However, he clarified that there are no known instances where AI itself autonomously initiated attacks. To address the risk of AI-related cyber attacks, Clark discussed Anthropic's measures, including a system called "safeguards" designed to limit the misuse of AI models, as well as the cybersecurity initiative "Project Glasswing." He emphasized the importance of collaborating with other organizations to enhance defense strategies and confirmed ongoing cooperation with the Japanese government and businesses.

Hidetoshi Tojo, the representative of Anthropic Japan, pointed out the challenges of utilizing AI in response to cyber attacks, specifically citing the need for speed in their approach. He remarked on the predictability of the increased cyber attacks, referencing discussions that began around the time of the spring 2026 release of the AI model Mythos.

Tojo also addressed misconceptions surrounding the AI model "Claude Mythos," which is sometimes viewed as the "only AI for defense." He stressed the necessity of leveraging other models for cyber attack responses, indicating a need for a more rapid and comprehensive approach in countermeasures.