OpenAI to provide latest cybersecurity AI model to Japanese firms
OpenAI to provide latest cybersecurity AI model to Japanese firms
https://www.asahi.com/ajw/articles/16584417
Publish Date: 2026-05-22 02:02:00
Source Domain: www.asahi.com
Using an unordered list, summarize the following article with between 4 and 8 key points.
The U.S.-based company that created ChatGPT plans to soon provide its latest artificial intelligence model–with enhanced cybersecurity capabilities and performance comparable to Claude Mythos–to some Japanese companies.
Jason Kwon, chief strategy officer of the major AI development firm OpenAI, revealed the plan in an interview with The Asahi Shimbun in Tokyo.
Anthropic, a U.S. startup, unveiled its new Claude Mythos model last month, which is known for its exceptionally advanced capability to identify and exploit system vulnerabilities.
Since then, there has been growing global concern over guarding against potential misuse of such advanced AI technologies.
OpenAI’s latest model is said to offer performance comparable to Mythos, and Japanese companies receiving it could potentially expand their range of defensive cybersecurity capabilities.
Kwon said his company intends to provide “GPT‑5.5‑Cyber,” a model specialized in cyberdefense, which OpenAI announced earlier this month, in response to Mythos.
GPT-5.5-Cyber has already been provided to government agencies and some companies in the United States.
“We’re trying to broaden access to Japan and other allies,” he said.
Last month, the British government’s AI Security Institute (AISI) found that GPT-5.5 had reached a performance level comparable to Mythos.
In a report released this month, AISI said that both models successfully cleared all tasks in simulated cyberattack trials.
Kwon said that Mythos “drew a lot of attention” from the media.
However, he added, “The progress of cyber capabilities actually has been continuous and there was not a single moment when things became very different,” expressing skepticism toward the belief that Mythos alone represented a sudden leap in performance.
AVOIDING LOSS OF CONTROL
Kwon also mentioned the roles of U.S. and Chinese companies leading AI development.
In 2023, then U.S. President Joe Biden and Chinese President Xi Jinping agreed to launch intergovernmental dialogue regarding AI development.
This month, U.S. Treasury Secretary Scott Bessent under the Trump administration also announced plans to begin discussions between the two nations on AI safety.
According to Kwon, in addition to government-level talks, companies and researchers from both countries continue discussions at the private-sector level.
“The United States and its allies are going to be in an intense competition with China and its allies for the foreseeable future,” Kwon said. “But it could still be done in a way where there’s still some mutual interest in making sure things don’t get out of hand.”
WORKING TOGETHER ON CYBERSECURITY
In 2024, OpenAI and Anthropic agreed with the U.S. government’s AISI to implement a framework for evaluating the safety of advanced AI models before their release.
This month, AISI’s successor organization reached similar agreements with Google LLC, Microsoft Corp. and xAI Corp.
A key challenge remains for how governments can govern powerful AI technologies, driven by private companies, which are increasingly difficult to control.
Kwon said OpenAI’s principle is to inform the government so that it can anticipate what will come next.
He added that “cybersecurity is a team sport,” and requires governments and all private-sector actors to work together.
Kwon also said that another OpenAI executive held discussions with Japanese government officials covering 15 critical infrastructure sectors, including finance and electricity, this week.
He added that the manufacturing sector could also be granted access.
