Reuters reported on Wednesday that Microsoft announced its intention to integrate Anthropic’s Claude Mythos Preview artificial intelligence model into its Security Development Lifecycle framework. This strategic step represents an official adoption of this powerful and unique AI technology model for discovering security vulnerabilities across the tech giant’s vast enterprise software portfolio.
With this innovative step, Microsoft becomes one of the first major technology companies globally to integrate this advanced model directly into its internal software engineering operations. Anthropic has described this pioneering model as currently outperforming any other artificial intelligence model by a wide margin in cyber capabilities. According to Reuters in its report, integrating the Claude Mythos model into the security development lifecycle will help engineers and developers identify complex security vulnerabilities and develop appropriate software fixes much faster at very early stages of the software development lifecycle, enhancing system resilience and reducing risks before products are released to users.
A hyper-powerful model shrouded in secrecy to avoid risks
This deep technical integration is a core part of Anthropic’s Project Glasswing, a restricted and exclusive initiative announced on April 7. This massive project brings together an elite group of technology giants and financial institutions, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, in an unprecedented collaborative effort aimed at securing critical software infrastructure globally.
In a related context, Anthropic confirmed that it does not plan to make the preview version of this model publicly available, pointing to the severe risks that could stem from adversaries and hackers utilizing its superior capabilities as a devastating cyber weapon. In a draft blog post published by mistake last March, the company clearly warned that this model “heralds an oncoming wave of AI models that can exploit vulnerabilities in ways that far outpace the efforts of defenders and cybersecurity officials.”
During pre-release testing phases, the model demonstrated terrifying capabilities, managing to identify thousands of zero-day security vulnerabilities, including critical flaws in every major operating system and every well-known web browser, according to Anthropic’s data. One of its most prominent and oldest discoveries was a 27-year-old software bug in OpenBSD, highlighting the extreme precision this system possesses.
In a statement to NBC News, Logan Graham, who leads offensive cyber research at Anthropic, said that the model is extremely advanced not only in its ability to uncover previously unknown software flaws, but also possesses the capacity to autonomously chain exploits together to break into complex systems and bypass multiple defenses.
Expanding the horizons of the strategic partnership between Microsoft and Anthropic
Microsoft has notably deepened its strategic partnership with Anthropic over recent months. In a bold forward-looking move in early 2026, the company made Anthropic models available as a default option across Microsoft 365 Copilot and other leading products, listing Anthropic as an official and trusted subprocessor within Microsoft’s enterprise data framework.
The partnership did not stop there; Microsoft also invested a massive $5 billion in Anthropic, securing in return a $30 billion commitment to use Azure cloud computing by the AI company, reflecting the scale of mutual trust and investments directed toward building a secure and efficient technological future.
Integrating the security development lifecycle and usage cost details
The integration of the security development lifecycle extends to reinforce this strategic relationship, making it an integral part of core engineering practices. It is worth noting that Microsoft’s Security Development Lifecycle framework, originally established as part of the company’s Trustworthy Computing initiative in 2002, is a mandatory and rigorous framework governing how all Microsoft software is designed, built, and tested. In February, Microsoft published expanded plans to update this framework to address AI security risks alongside traditional software concerns.
For its part, Anthropic committed to providing $100 million in model usage credits to cover costs for participants in Project Glasswing. After the research preview phase concludes, the model will be available to participants at a specified cost of $25 per million input tokens and $125 per million output tokens. This advanced access will be provided through the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Frequently Asked Questions
What is the new AI model and why will it not be made public?
It is an ultra-advanced AI model from Anthropic designed to discover complex security vulnerabilities. It will not be made available to the general public to avoid security risks, as the company fears hackers could use its superior capabilities as a weapon to breach systems rather than protect them.
What companies are participating in Project Glasswing?
The project includes an alliance of major tech and financial companies, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks.
How will Microsoft benefit from integrating this model?
Integrating the model into Microsoft’s security development lifecycle will help identify security vulnerabilities and develop fixes faster in the early stages of software development, raising the security level of systems before they reach users.
