Jack Clark, Anthropic co-founder and head of its public benefit division, said on October 14 that AI may have been used in a series of cyberattacks on Japanese companies. Speaking at a media briefing marking the first anniversary of Anthropic Japan, he said hackers could be misusing publicly available open-weight models (models whose trained weights are available for users to download). Details of the attacks remain unclear, but Clark said AI tools may have been used in some way.

Clark said that, to his knowledge, there were no confirmed cases of AI itself autonomously launching attacks. The possibility that AI was used in attacks and whether AI independently carried them out were presented as separate questions.

Anthropic cited safeguards designed to limit misuse of its AI models and its cybersecurity project, Project Glasswing, as part of its response. Clark stressed the importance of working with other organizations on defense and said Anthropic would continue cooperating with Japan’s government and businesses.

Hidetoshi Tojo, head of Anthropic Japan, identified the pace of response as a challenge for Japanese companies using AI to counter cyberattacks. He said the run of attacks had been somewhat predictable, adding that the topic had been discussed since around spring 2026, when Claude Mythos appeared.

Tojo said some people mistakenly view Claude Mythos as the “only AI used for defense.” He said other models should also be used to counter cyberattacks and that responses need to move quickly. The report did not provide details or specific cases showing whether AI was used in the attacks. It said Clark had raised the possibility of misuse, including of open-weight models, while Anthropic had pointed to defensive measures and cooperation between government and industry. It also did not identify any model used in the attacks or specify concrete cooperation plans or implementation dates. Tojo’s reference to the “only AI used for defense” was an example of a misconception; he argued that companies should use other AI models rather than rely on a single one. The briefing was held for Anthropic Japan’s first anniversary, and Clark said the company intended to continue working with Japan’s government and businesses.

Jack Clark, Anthropic co-founder and head of its public benefit division, said on October 14 that AI may have been used in a series of cyberattacks on Japanese companies. Speaking at a media briefing marking the first anniversary of Anthropic Japan, he said hackers could be misusing publicly available open-weight models (models whose trained weights are available for users to download). Details of the attacks remain unclear, but Clark said AI tools may have been used in some way.

Clark said that, to his knowledge, there were no confirmed cases of AI itself autonomously launching attacks. The possibility that AI was used in attacks and whether AI independently carried them out were presented as separate questions.

Anthropic cited safeguards designed to limit misuse of its AI models and its cybersecurity project, Project Glasswing, as part of its response. Clark stressed the importance of working with other organizations on defense and said Anthropic would continue cooperating with Japan’s government and businesses.

Hidetoshi Tojo, head of Anthropic Japan, identified the pace of response as a challenge for Japanese companies using AI to counter cyberattacks. He said the run of attacks had been somewhat predictable, adding that the topic had been discussed since around spring 2026, when Claude Mythos appeared.

Tojo said some people mistakenly view Claude Mythos as the “only AI used for defense.” He said other models should also be used to counter cyberattacks and that responses need to move quickly. The report did not provide details or specific cases showing whether AI was used in the attacks. It said Clark had raised the possibility of misuse, including of open-weight models, while Anthropic had pointed to defensive measures and cooperation between government and industry. It also did not identify any model used in the attacks or specify concrete cooperation plans or implementation dates. Tojo’s reference to the “only AI used for defense” was an example of a misconception; he argued that companies should use other AI models rather than rely on a single one. The briefing was held for Anthropic Japan’s first anniversary, and Clark said the company intended to continue working with Japan’s government and businesses.