パロアルトのセキュリティ企業が、Anthropicの「Claude Mythos Preview」を活用し、わずか5日間でmacOSの有効なエクスプロイトを構築したと発表しました。研究チームはApple Parkを訪問し、Appleの担当者に対して調査結果を共有しました。
カリフォルニア州パロアルトに拠点を置くCalifの研究者らは、Anthropicの「Claude Mythos Preview」の支援を受けて、macOS向けの権限昇格エクスプロイトを設計しました。このツールは既知の脆弱性クラスを迅速に特定しましたが、コードを完成させるには依然として人間の専門知識が不可欠でした。作成されたエクスプロイトにより、通常は制限されているMacBookの領域へアクセスが可能となり、M5チップを搭載したデバイスにおいて攻撃者が制御権を奪取できることが確認されました。
AIによるレポート AIによって生成された画像
Anthropic has limited access to its Claude Mythos Preview AI model due to its superior ability to detect and exploit software vulnerabilities, while launching Project Glasswing—a consortium with over 45 tech firms including Apple, Google, and Microsoft—to collaboratively patch flaws and bolster defenses. The announcement follows recent data leaks at the firm.
Anthropic has released a new cyber-focused AI model called Mythos, capable of detecting software flaws faster than humans and generating exploits. The model has raised alarms among governments and companies for potentially turbocharging hacking by exposing vulnerabilities quicker than they can be patched. Officials worldwide are scrambling to assess the risks.
AIによるレポート
Following last week's unveiling that sparked global alarms, Anthropic has restricted its powerful Mythos AI—adept at finding cybersecurity vulnerabilities—to select firms under Project Glasswing, including Amazon Web Services, Apple, and Google, after an accidental leak raised national security concerns.
The Trump administration directed Anthropic to revoke access to its advanced AI models following concerns about ties to China. The move led the company to take its most powerful offerings offline last week.
AIによるレポート
The White House has permitted Anthropic to provide its advanced Claude Mythos 5 model to more than 100 US organizations. The decision follows weeks of negotiations between the company and government officials.
Anthropic has introduced Claude Design, a new research preview tool that enables subscribers to generate designs, prototypes and slides using its Claude AI. Powered by the Opus 4.7 vision model, the tool emphasizes workplace visuals over arbitrary image generation. It is available now to Pro, Max, Team and Enterprise subscribers.
AIによるレポート
Anthropic has restricted unlimited access to its Claude AI models through third-party agents like OpenClaw, requiring heavy users to pay extra via API keys or usage bundles starting April 4, 2026. The policy shift, announced over the weekend, addresses severe system strain from high-volume agent tools previously covered under $20 monthly subscriptions.