WayToClawEarn
High impactAnthropic 官方

Anthropic releases Project Glasswing: a century alliance that uses AI to protect network security

Anthropic joined forces with technology giants such as AWS, Apple, Google, and Microsoft to launch Project Glasswing, which uses the Claude Mythos Preview model to automatically discover zero-day vulnerabilities in critical software. The model has autonomously discovered thousands of high-severity vulnerabilities, including a remote crash vulnerability that had gone undiscovered in OpenBSD for 27 years. Anthropic committed $100 million in model credits.

WayToClawEarn EditorialPublished May 23, 2026Updated Aug 8, 2026

Editorial review of public sources · AI-assisted drafting. How we work · Original source

Core conclusion

On May 22, 2026, Anthropic officially released Project Glasswing, a cybersecurity initiative jointly participated by technology giants such as AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, etc., aiming to use cutting-edge AI models to build a defense system for the world's critical software infrastructure.

The core model Claude Mythos Preview (a cutting-edge model that has not been publicly released) demonstrates code security capabilities that shock the industry: it has independently discovered thousands of high-risk zero-day vulnerabilities, covering all major operating systems and browsers. These include:

NumberVulnerability DiscoverySeveritySignificance
1OpenBSD unfixed vulnerability for 27 years⚠️ High riskRemote connection can crash the system
2FFmpeg 16-year legacy vulnerability⚠️ High riskUndiscovered by automated testing 5 million times
3Chain exploitation of multiple vulnerabilities in the Linux kernel⚠️ SevereElevate ordinary user rights to full control

Key Points

  • Published: May 22, 2026
  • Affects: All teams that rely on AI tools for code development and content production
  • Core Change: The security capabilities of AI models have surpassed top human security experts, which means that the AI toolchain itself also requires more stringent security audits
  • Anthropic Commitment: $100 million in Mythos Preview model usage + $4 million in open source security donations

Background: Cybersecurity Crisis in the AI Era

Annual global cybercrime losses currently cost approximately $500 billion. Many critical software vulnerabilities remain undiscovered for years or even decades because finding and exploiting them requires highly specialized skills. With the explosion of cutting-edge AI model capabilities, this balance is being disrupted.

Claude Mythos Preview significantly outperforms the previous generation model Opus 4.6 in multiple security benchmarks:

BenchmarksMythos PreviewOpus 4.6Improvements
CyberGym (Vulnerability Reproduction)83.1%66.6%+16.5%
SWE-bench Verified93.9%80.8%+13.1%
SWE-bench Pro77.8%53.4%+24.4%
Terminal-Bench 2.082.0%65.4%+16.6%

Even more amazingly, Mythos Preview works completely autonomously—no human guidance, no prompt word engineering. It's like an "automatic vulnerability digger".

Core participating companies and roles

Project Glasswing’s partners cover the entire technology industry chain:

Company TypeRepresentative CompanyRole
Cloud ComputingAWS, Google Cloud, Microsoft AzureProviding inference infrastructure
ChipApple, Broadcom, NVIDIAHardware security hardening
SecurityCrowdStrike, Cisco, Palo Alto NetworksSecurity Practice Integration
FinanceJPMorganChaseFinancial System Security
Open SourceLinux FoundationOpen Source Ecological Security

Impact on AI content producers

What does this news mean for your AI workflow?

Short term (next 1-3 months):

  • If you use Claude Code or Claude API for automated content production, the security capabilities of the Claude series models are rapidly improving. Anthropic plans to build Mythos-level security into next Claude Opus release
  • Security audit steps introduced in AI Agent automation processes will become standard

Medium term (3-6 months):

  • Automated code audit and vulnerability scanning tools will be fully AI-based
  • Open source maintainers will receive free AI security audit capabilities

Long term (6-12 months):

  • AI security will become an independent track, and leading AI tool manufacturers will launch their own security protection solutions
  • Using AI tools without security audits will lead to higher supply chain security risks

Tool entry paragraph

The hero of this news is Claude (specifically Claude Mythos Preview), an unreleased cutting-edge model developed by Anthropic. If you are using Claude Code or ChatGPT of OpenAI for automated content production, it is crucial to understand the significance of this leap in security capabilities - in the future, your AI tool chain will not only need to be able to write code, but also need to be able to discover and repair security vulnerabilities independently. Competitors such as DeepSeek and Gemini will most likely follow up with similar security capability building.

Claude Mythos Preview

Related extended information

Internal link guidance

View source →

Disclaimer: this site shares educational insights only, for inspiration and reference. No outcome guarantee; external execution and decisions are your own responsibility.