AI Model Mythos: Companies Using It Describe It as Having "Game-Changing Potential"

Recently NHK News on-line reported the following:

AIモデル ミュトス 利用企業 “ゲームチェンジ起こす能力”

2026年5月30日15:42

生成AI・人工知能

性能が非常に高くサイバー攻撃への悪用も懸念されるAIモデル、「クロード・ミュトス」の提供を受けたアメリカの企業がNHKのインタビューに応じ、その性能について「ゲームチェンジを起こす能力だ」と証言しました。そのうえで、「防御側の手にあり、攻撃側の手に渡っていない今この瞬間が重要だ」として悪用される前に対策が必要だと訴えました。

アメリカの新興AI企業、アンソロピックが開発したAIモデル「ミュトス」は、ソフトウエアのぜい弱性を発見する能力が極めて高いとされ、会社はサイバー攻撃に悪用されるおそれもあるとして公開を見送った一方、先月からIT大手や金融機関などおよそ50の企業や組織に提供してサイバー攻撃に備えるプロジェクトを始めています。

このプロジェクトに参加して「ミュトス」の提供を受けたアメリカ企業「モジラ」でウェブブラウザーの技術責任者を務めるボビー・ホーリー氏が今月下旬、NHKのインタビューに応じました。

この中でホーリー氏は、「ミュトス」の性能について、「1つの欠陥を見つけるために動ける一流の研究者は限られている一方で、AIなら数百、あるいは数千もの分析を並行して実行できる。これはセキュリティーの世界において、ゲームチェンジを起こす能力だ」と述べ、衝撃を受けたと証言しました。

また、ホーリー氏は、「ミュトス」の活用などによってこの会社が4月にウェブブラウザーのぜい弱性を修正した件数は平均的な1か月間の修正件数のおよそ20倍に増加したと明らかにしました。

そのうえで、「今これらの最先端の能力は防御側の手にある。攻撃側の手に渡っていない今この瞬間を活用し、私たち全員が依存しているインターネットのインフラを強化することが極めて重要だ」と述べ、悪用される前に対策が必要だと訴えました。

Translation

AI Model Mythos: Companies Using It Describe It as Having "Game-Changing Potential"

May 30, 2026, 15:42

Generative AI/Artificial Intelligence

An American company that received the AI ​​model "Claude Mythos," which boasted extremely high performance and raised concerns about its potential misuse in cyberattacks, gave an interview to NHK, testifying that its performance "has the potential to change the game." Furthermore, they emphasized that " at this very moment, the crucial thing is that it's in the hands of the defenders and not yet in the hands of the attackers,” appealing for countermeasures before being misused.

The AI ​​model "Mythos," developed by the American startup AI company Anthropic, was said to have an extremely high ability to detect software vulnerabilities. While the company withheld its public release due to concerns about its potential misuse in cyberattacks, it began a project last month to provide it to approximately 50 companies and organizations, including major IT companies and financial institutions, to prepare for cyberattacks.

Bobby Hawley, the Chief Technology Officer for web browsers at Mozilla as an American company that received "Mythos" as part of this project, gave an interview to NHK late this month.

In his remarks, Mr. Hawley described the capabilities of "Mythos," stated that "While only a limited number of top-tier researchers can work to find a single software vulnerability, AI can perform hundreds, or even thousands, of analyses in parallel. This is a game-changing capability in the world of security," and testified that he was shocked by this.

Mr. Hawley also revealed that, thanks to the use of "Mythos," the company's number of web browser vulnerabilities fixed in April increased to approximately 20 times the average number of fixes in a single month.

On top of that he said, "These cutting-edge capabilities are now in the hands of the defenders. It is crucial to made use of the present moment to strengthen the internet infrastructure on which we all depend on, before they fall into the hands of attackers”, urging that countermeasures were needed before someone used it with an ill intention.

So, an American company’s AI model "Claude Mythos," which is said to have extremely high performance, could be a game changer. Apparently, it is crucial that this AI model will not go into the hands of someone who may use it with ill intent.