Security · October 17, 2026
Anthropic Co-Founder Discusses Cybersecurity and AI Tools
On October 14, 2026, at a media briefing held by the Japanese branch of Anthropic, co-founder Jack Clark addressed the potential use of artificial intelligence tools in the context of recent cyberattacks. Clark, who was visiting Japan for the first time, speculated that AI tools could indeed be employed by hackers, stating, "It is likely that AI tools are being used." This statement came as he responded to questions regarding the relationship between cyberattacks and AI during the event.
Clark highlighted that hackers, much like companies, utilize available tools, suggesting that AI's involvement in such attacks is a possibility. He explained that Anthropic is implementing measures to prevent misuse through initiatives like "safeguards" and "Project Glasswing," which prioritizes the delivery of the latest AI models to enterprises. However, he noted that the existence of other models and open-weight models means that Anthropic alone cannot resolve the situation.
Emphasizing the importance of collective action, Clark remarked, "It is insufficient for just Anthropic to act responsibly. The entire ecosystem must address this issue. Therefore, we dedicate time to policy discussions aimed at creating sensible traffic rules that apply equally to all operators and foster a safer environment for everyone."
When asked by reporters whether there have been instances of AI autonomously launching attacks, Clark denied such occurrences, stating, "We have not confirmed any such instances."
Clark also touched upon the concept of "recursive self-improvement" (RSI) in AI, suggesting that this capability might become possible as early as 2027. He referenced his newsletter, which notes evidence of AI systems beginning to demonstrate creativity in scientific tasks ranging from mathematics to biology. He argued that the field of AI research is a form of science itself and urged that advancements in AI-driven science could progress at a rapid pace next year.
Also present at the event was Hidetoshi Tojo, the president of Anthropic's Japanese branch, who echoed Clark's concerns regarding the surge in cyberattacks. He pointed out that the emergence of Mythos in spring had made the situation predictable, but he added, "There are still actions that can be taken now," encouraging the use of AI to address these challenges.