Project Glasswing
In plain wordsAn Anthropic program, launched in 2026, that gave a hacking-capable AI model only to organizations defending important software instead of releasing it to everyone.
An industry initiative led by AnthropicAnthropic, prop. n.: An American AI company that makes the Claude chatbot and models. It was founded in 2021 by former OpenAI staff and puts particular emphasis on AI safety research. that gives restricted access to Claude Mythos Preview — a model the company judged too capable at finding and exploiting software vulnerabilities for general release — to organizations that maintain critical software, so they can find and fix flaws before attackers do. Launch partners included Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia and Palo Alto Networks, and access was extended to more than 40 further organizations. Anthropic committed up to US$100 million in usage credits and US$4 million in donations to open-source security groups, and reported that the model had already turned up thousands of serious vulnerabilities, affecting every major operating system and web browser. It is a prominent example of defender-first staged deploymentstaged deployment, n.: Releasing a new AI system step by step — first to testers, then to limited groups, then widely — so that problems can be spotted and fixed before everyone has it. of a model with dangerous capabilities.
Our maintainers got Mythos access through Project Glasswing and patched three decades-old parser bugs.
Origin
Announced by Anthropic on 7 April 2026 together with Claude Mythos Preview. Access to Anthropic's later models Claude Fable 5 and Claude Mythos 5 was suspended from 12 June 2026 under a US government export control directiveexport control directive, n.: A US government order, used against AI models in June 2026, telling a company to stop foreigners from using some of its AI models for national-security reasons. that was lifted at the end of that month.
Usage note
Capability figures are Anthropic's own and were not independently verified at launch.
Related
- staged deploymentstaged deployment, n.: Releasing a new AI system step by step — first to testers, then to limited groups, then widely — so that problems can be spotted and fixed before everyone has it.
- AI-assisted vulnerability discoveryAI-assisted vulnerability discovery, n.: Using AI to find security holes in software — and often to suggest fixes — faster than human experts alone, so that defenders, or attackers, can act on them.
- dual-usedual-use, adj.: Describes technology or knowledge that can be used both to help and to harm: the same AI that finds software bugs for defenders can find them for attackers.
- export control directiveexport control directive, n.: A US government order, used against AI models in June 2026, telling a company to stop foreigners from using some of its AI models for national-security reasons.
- ExploitGymExploitGym, prop. n.: A test that measures whether AI agents can turn a known software security flaw into a working attack, using hundreds of real vulnerabilities.
- AnthropicAnthropic, prop. n.: An American AI company that makes the Claude chatbot and models. It was founded in 2021 by former OpenAI staff and puts particular emphasis on AI safety research.
Sources
- Constellation Research (2026). Anthropic launches Project Glasswing, Claude Mythos Preview in AI cybersecurity push. 7 April 2026. constellationr.com/insights/news/anthropic-launches-project-glasswing-claude-mythos-preview-ai-cybersecurity-push
- VentureBeat (2026). Anthropic says its most powerful AI cyber model is too dangerous to release. April 2026. venturebeat.com/data/anthropic-says-its-most-powerful-ai-cyber-model-is-too-dangerous-to-release
Mentioned in
export control directiveexport control directive, n.: A US government order, used against AI models in June 2026, telling a company to stop foreigners from using some of its AI models for national-security reasons.