{"slug": "anthropic-puts-its-most-restricted-model-claude-mythos-5-into-claude-security", "title": "Anthropic Puts Its Most Restricted Model Claude Mythos 5 Into Claude Security", "summary": "Anthropic has integrated Claude Mythos 5, its restricted cyber model, into Claude Security for Enterprise customers, enabling vulnerability scanning with CWE classification, confidence and severity ratings, and suggested fixes, billed as standard token usage under existing plans. The move follows Project Glasswing, which found over 10,000 high- or critical-severity flaws across 50 partners, and includes a $35 million Defender Advantage Fund for open-source security work.", "body_md": "*Anthropic has put Claude Mythos 5, its restricted cyber model, inside Claude Security for Enterprise customers. The bet is simple: give defenders the results without handing everyone the raw model.*\n\nAnthropic made the move on August 21, 2026, and the details matter if you run security inside a company that already pays for Claude Enterprise. Claude Security can scan a repository, trace data flows across files, and return vulnerability findings with a CWE category, confidence and severity ratings, and a suggested fix for human review. According to Anthropic's launch post, those Mythos 5 scans are billed as standard token usage under the existing Enterprise plan. No separate add-on.\n\nThat's the commercial hook.\n\nSecurity tooling usually lives in its own budget fight, with its own vendor review and renewal cycle. Anthropic is trying to make AI vulnerability scanning feel less like a new platform purchase and more like turning on a capability customers already have access to. For security teams, that distinction isn't cosmetic. It can be the difference between testing a tool this quarter and waiting for procurement to catch up next year.\n\n## From Glasswing to Claude Security\n\nClaude Mythos 5 didn't arrive in Claude Security from nowhere. In April, Anthropic launched Project Glasswing with Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks, and others to scan critical software with Claude Mythos Preview. Anthropic later said the first roughly 50 partners found more than 10,000 high- or critical-severity flaws, then expanded the program in June to about 150 additional organizations across more than 15 countries.\n\n[Hugging Face's CEO says being labeled too dangerous is the best enterprise marketing an AI lab can get](https://startupfortune.com/hugging-faces-ceo-says-being-labeled-too-dangerous-is-the-best-enterprise-marketing-an-ai-lab-can-get/)\n\nHugging Face CEO Clem Delangue publicly challenged the 'too dangerous to release' label on Anthropic's Mythos model on June 29, arguing it functions more as enterprise marketing than a genuine safety signal. With Anthropic filing its S-1 amid a restricted rollout of Mythos under Project Glasswing, the disagreement maps a real fault line: whether... - [dangerous AI models enterprise sales](https://startupfortune.com/hugging-faces-ceo-says-being-labeled-too-dangerous-is-the-best-enterprise-marketing-an-ai-lab-can-get/) - [restricted models CSO security value](https://startupfortune.com/hugging-faces-ceo-says-being-labeled-too-dangerous-is-the-best-enterprise-marketing-an-ai-lab-can-get/)\n\nThat history explains the caution here. Glasswing was gated: organizations had to meet Anthropic's security requirements and use the model for defensive work. Claude Security keeps that idea but changes the shape of the product. No free-form Mythos 5 prompt box for cyber work. Instead, users get a scan result, a location, a classification, and a proposed patch.\n\nA human still has to approve the fix before it ships. Full stop.\n\nThat control is the product. Anthropic is saying the dangerous part is direct access to a model that can reason deeply about vulnerabilities, so it is packaging the output instead. You can argue with the safety logic, but the business logic is clear enough. Banks, cloud providers, software vendors, and infrastructure operators want the capability. They also want someone else to explain why it won't become a new risk sitting inside the codebase.\n\nAnthropic is adding money to that argument through the Defender Advantage Fund, or 0xDAF. The company says the fund will provide $35 million in Claude credits for organizations working on open-source security, including groups patching live vulnerabilities in widely used projects and teams building repeatable scanning and patching workflows. Initial recipients haven't been named yet. That dry detail is worth keeping because it tells you where the announcement still has an open tab.\n\n## The Enterprise Test\n\nOpenAI is coming at the same broad coding market from a different direction. In June, OpenAI said Codex had more than 5 million weekly active users, up more than sixfold since the launch of its desktop app in February, and that knowledge workers had grown to about 20% of the user base. OpenAI is chasing reach. Anthropic is chasing trust from the institutions that need AI to be useful and contained at the same time.\n\nFrankly, that's the harder sale and the cleaner one. A coding assistant can win by becoming habit-forming across millions of workers. A security product wins only when a compliance team, a CISO, and a budget owner can all live with the same answer. Anthropic spent months using Glasswing to prove that Mythos-level scanning could find serious bugs before offering it more broadly through Claude Security.\n\nRivals won't sit still. Google and Microsoft both have AI-assisted security work of their own. Microsoft was already named as a Glasswing launch partner. The issue isn't whether another lab can build a scanner that reads code and suggests fixes. The issue is whether enterprise security teams will trust a frontier model near production code before an attacker uses a similar tool against them.\n\n[Europe wants access to Anthropic's Mythos before AI risk widens](https://startupfortune.com/europe-wants-access-to-anthropics-mythos-before-ai-risk-widens/)\n\nThe European Commission is pressing Anthropic for closer access to Claude Mythos Preview, the unreleased AI model behind Project Glasswing. The talks show how frontier model access is becoming a strategic lever in cybersecurity, regulation and enterprise AI adoption. - [how to access Anthropic Claude Mythos Preview model](https://startupfortune.com/europe-wants-access-to-anthropics-mythos-before-ai-risk-widens/) - [European Commission AI cybersecurity access requirements](https://startupfortune.com/europe-wants-access-to-anthropics-mythos-before-ai-risk-widens/)\n\nAnthropic's answer is controlled access, human review, and billing through a plan companies already bought. That won't settle the AI security race. It does give customers a practical way to test whether Mythos 5 is more useful as a locked-down defender than as another impressive model sitting behind a waiting list.\n\n**Also read:** [OpenAI Cuts GPT-5.6 Sol API Prices After Holding the Line for Months](https://startupfortune.com/openai-cuts-gpt-56-sol-api-prices-after-holding-the-line-for-months/) • [Nvidia Shows a Better AI Harness Beat a Smarter Model on ARC-AGI-3](https://startupfortune.com/nvidia-shows-a-better-ai-harness-beat-a-smarter-model-on-arc-agi-3/) • [Dutch Regulator Fines Uber Nearly $1 Billion Over Automated Driver Suspensions](https://startupfortune.com/dutch-regulator-fines-uber-nearly-1-billion-over-automated-driver-suspensions/)", "url": "https://wpnews.pro/news/anthropic-puts-its-most-restricted-model-claude-mythos-5-into-claude-security", "canonical_source": "https://startupfortune.com/anthropic-puts-its-most-restricted-model-claude-mythos-5-into-claude-security/", "published_at": "2026-08-22 06:14:39+00:00", "updated_at": "2026-08-22 06:42:24.005862+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-safety", "ai-policy"], "entities": ["Anthropic", "Claude Mythos 5", "Claude Security", "Project Glasswing", "Amazon Web Services", "Apple", "Broadcom", "Cisco"], "alternates": {"html": "https://wpnews.pro/news/anthropic-puts-its-most-restricted-model-claude-mythos-5-into-claude-security", "markdown": "https://wpnews.pro/news/anthropic-puts-its-most-restricted-model-claude-mythos-5-into-claude-security.md", "text": "https://wpnews.pro/news/anthropic-puts-its-most-restricted-model-claude-mythos-5-into-claude-security.txt", "jsonld": "https://wpnews.pro/news/anthropic-puts-its-most-restricted-model-claude-mythos-5-into-claude-security.jsonld"}}