Anthropic Launches Opus 4.7 and Restricted Mythos AI for Cybersecurity
Anthropic released Claude Opus 4.7, its most powerful general-purpose AI model, while simultaneously launching Project Glasswing. This initiative restricts the more capable Mythos model to select partners like Microsoft and Apple for defensive cybersecurity operations. Due to fears that Mythos could be weaponized for hacking, Anthropic withheld public access, instead providing resources to patch critical infrastructure vulnerabilities. The dual strategy balances commercial AI advancement with urgent security needs, marking a significant shift in how the tech industry addresses AI-driven cyberthreats and market dynamics.
Cross-source coverage
Wire timeline
Anthropic Releases Claude Opus 4.7, Retaking Lead in Powerful LLM Market
Anthropic has publicly released Claude Opus 4.7, its most powerful large language model to date, narrowly surpassing rivals OpenAI's GPT-5.4 and Google's Gemini 3.1 Pro in key benchmarks. The new model excels in agentic coding, scaled tool-use, and financial analysis, achieving a leading Elo score of 1753 on the GDPVal-AA knowledge work evaluation. While it does not dominate every category, losing ground in agentic search and multilingual Q&A, Opus 4.7 is optimized for reliability and long-horizon autonomy. Significant architectural upgrades include high-resolution multimodal support, allowing image processing up to 3.75 megapixels, which drastically improves visual acuity for computer-use agents. The model also features enhanced 'rigor,' enabling autonomous self-correction to reduce hallucinations. Claude Opus 4.7 is available immediately across major cloud platforms like Amazon Bedrock, Google Cloud, and Microsoft Foundry, with API pricing remaining steady at $5/$25 per million tokens. Meanwhile, Anthropic continues to restrict its even more advanced model, Mythos, to select enterprise partners for cybersecurity testing.
VentureBeatAnthropic Releases Claude Opus 4.7 Amid Mythos Preview Buzz
Anthropic has officially launched Claude Opus 4.7, described as its most powerful generally available AI model to date. This new iteration improves upon Opus 4.6 in advanced software engineering, complex coding tasks, image analysis, and instruction following. However, the release occurs alongside significant attention on Anthropic's more capable but restricted Mythos Preview model, which outperforms Opus 4.7 in all evaluations. Due to security concerns, Mythos Preview remains limited to select partners like Nvidia and Microsoft, while Opus 4.7 serves as a testing ground for new cybersecurity safeguards. Anthropic intentionally reduced certain cyber capabilities in Opus 4.7 compared to Mythos to ensure safer broad deployment. The company aims to use findings from this release to facilitate the eventual public launch of Mythos-class models. Security professionals can access enhanced features through the new Cyber Verification Program. Major tech customers including Intuit, Shopify, and Databricks participated in early testing. Pricing remains unchanged at $5 per million input tokens and $25 per million output tokens, maintaining continuity with the previous version while offering incremental improvements for general enterprise use.
The VergeAnthropic Launches Claude Opus 4.7, a Safer but Less Capable AI Model
Anthropic has officially released Claude Opus 4.7, its most powerful generally available artificial intelligence model to date. While the company states that Opus 4.7 represents an improvement over previous versions like Opus 4.6 in areas such as software engineering, instruction following, and agentic computer use, it is explicitly described as 'less broadly capable' than the specialized Claude Mythos Preview. The Mythos model, recently introduced under the 'Project Glasswing' cybersecurity initiative, possesses advanced capabilities for identifying software security flaws but remains restricted to a select group of corporate partners due to high-risk potential. Opus 4.7 incorporates automated safeguards designed to detect and block prohibited cybersecurity requests, reflecting Anthropic's strategy to prioritize safe deployment. This release follows significant discussions involving the Trump administration and tech leaders regarding AI security risks. The new model is available via Anthropic’s API and major cloud providers including Microsoft, Google, and Amazon at the same price point as its predecessor, serving as a step toward the eventual scaled deployment of Mythos-class technologies.
US Top News and AnalysisAnthropic Unveils Opus 4.5 AI Model with Major Performance Upgrades
Anthropic has officially launched Opus 4.5, the final release in its 4.5 series of artificial intelligence models, marking a significant advancement in coding, tool use, and general reasoning capabilities. The new model achieves state-of-the-art results on key benchmarks, becoming the first to score above 80% on SWE-Bench Verified for code generation accuracy. Alongside the model release, Anthropic is expanding access to its productivity tools, including Claude for Chrome and Claude for Excel, available to various subscriber tiers. A notable feature of Opus 4.5 is its enhanced long-context performance and memory management, enabling an 'endless chat' function that automatically compresses earlier conversation parts without user notification. This update is designed to support complex agentic workflows where Opus orchestrates tasks handled by smaller models like Haiku. The launch positions Anthropic competitively against recent releases from rivals, including OpenAI's GPT 5.1 and Google's Gemini 3. The announcement highlights substantial gains in working-memory capabilities, addressing the need for AI systems to retain relevant details over extended interactions.
thenationalnewsAnthropic Withholds 'Claude Mythos' AI to Prevent Widespread Hacking
Anthropic has decided against public release of its latest AI model, Claude Mythos, after it demonstrated an unprecedented ability to expose thousands of software vulnerabilities in common applications. The model identified flaws dating back 27 years that had evaded detection by human developers and rigorous testing. Fearing the tool could be weaponized for widespread hacking if released broadly, Anthropic is instead collaborating with cybersecurity specialists through a project called 'Glasswing.' This alliance includes major tech firms like Amazon, Apple, Microsoft, Cisco, and Palo Alto Networks, as well as the Linux Foundation. Anthropic is providing $100 million in computing resources to help these partners use Mythos defensively to patch critical infrastructure before adversaries can exploit the weaknesses. The company notes that the window between vulnerability discovery and exploitation has collapsed from months to minutes due to AI capabilities.
The GuardianAnthropic Unveils Project Glasswing to Transform Cybersecurity Defense
On April 8, 2026, AI giant Anthropic announced the launch of Project Glasswing, a strategic initiative designed to integrate its advanced Claude Mythos2 Preview model into defensive cybersecurity operations. This partnership includes major industry players such as Microsoft, Apple, CrowdStrike, Palo Alto Networks, and JPMorgan Chase. The announcement follows reports that the new model has identified vulnerabilities across all major operating systems and web browsers. While aimed at bolstering digital defenses, the move raises significant questions about the future viability of traditional cybersecurity firms in an era where AI can autonomously detect and potentially exploit system flaws. The initiative highlights a critical shift in the technology sector, prompting a reevaluation of how security industries adapt to or survive against powerful generative AI tools.
Technology - WSJ.comAnthropic Unveils Project Glasswing to Integrate AI into Cybersecurity Defense
Anthropic has launched Project Glasswing, a new initiative partnering with major cybersecurity firms and tech giants to leverage its Claude Mythos2 Preview model for defensive security operations. The project aims to address critical vulnerabilities found in all major operating systems and web browsers by providing select participants, including CrowdStrike, Microsoft, Apple, and Google, with advanced AI tools. This move comes amidst industry concerns following a February stock selloff, prompting questions about the future viability of traditional cybersecurity sectors in the age of artificial intelligence. By integrating frontier AI models directly into defense strategies, Anthropic signals a significant shift in how organizations approach threat detection and mitigation, potentially disrupting existing market dynamics.
WSJ.com: US BusinessAnthropic to Preview 'Mythos' AI Model for Cybersecurity Defense
Anthropic is preparing to release a preview of its new artificial intelligence model, named Mythos, designed to help major technology companies identify and patch vulnerabilities in their hardware and software. The company will provide access to this advanced tool to approximately 50 key organizations responsible for maintaining critical infrastructure. Recipients include industry giants such as Amazon, Microsoft, Apple, Alphabet-owned Google, and the Linux Foundation. This initiative aims to proactively ward off emerging AI-driven cyberthreats by empowering these entities with sophisticated detection capabilities. The move highlights Anthropic's strategic focus on enhancing global digital security through collaborative AI development.
Technology - WSJ.com