Microsoft Challenges Anthropic with Its Half-Priced Cybersecurity AI

Microsoft recently unveiled an innovative artificial intelligence model for cybersecurity, named MAI-Cyber-1-Flash. This model, in conjunction with the MDASH security system and OpenAI's GPT-5.4 model, outperforms Anthropic's Claude Mythos 5 model by 12 points on a key performance metric. Microsoft designed this solution to leverage AI in defending against other AIs.

The project, dubbed Project Perception, combines MAI-Cyber-1-Flash and MDASH, the latter of which was launched in May. This combination is set to enter public preview on August 3, and will be directly integrated into Microsoft Defender. Microsoft plans a gradual rollout of this technology across its security products.

The benchmark results published by Microsoft are impressive. The combination of these technologies achieved a score of 96% on the CyberGym platform, surpassing the 84% score obtained by the Mythos 5 model. This result demonstrates the enhanced effectiveness of Microsoft's solution compared to its competitors.

In terms of pricing, Microsoft has opted for a consumption-based model, measured by the number of security compute units (SCUs) used. The more AI agents execute scenarios, the more SCUs they consume, which increases costs. However, Microsoft claims that this new configuration allows for savings of nearly 50% compared to current solutions based on MDASH.

During a presentation, Mustafa Suleyman detailed the transfer process between MAI-Cyber-1-Flash and GPT-5.4. According to him, MAI-Cyber-1-Flash is capable of handling about 90% of queries. The remaining 10% of queries are transferred to GPT-5.4, a model approximately 10 times larger, for resolution. This collaboration between the models not only improves overall performance but also halves costs.

Suleyman described the results achieved on CyberGym as "truly remarkable," emphasizing the significance of this technological advancement.

This development comes shortly after the launch of Anthropic's Claude Fable 5, the first model in the Mythos series to be made public. Anthropic had initially touted the effectiveness of Mythos in detecting cybersecurity vulnerabilities, to the extent of fearing it could "break the Internet" if misused. This situation led to a rollback on the launches of Fable 5 and Mythos 5, particularly after the U.S. government discovered a method to bypass the model's limitations. At its initial announcement, Mythos had been released only to selected government agencies and technology professionals.

In conclusion, Mustafa Suleyman highlighted Microsoft's long experience as a guardian of sensitive data for governments and businesses worldwide. This expertise, combined with a vast amount of accumulated data, has enabled Microsoft to develop this advanced cybersecurity model. With the expertise of its cybersecurity specialists, Microsoft positions itself as a major player in the field of digital security.