Microsoft is reportedly preparing a new security product, called Project Perception, that would work similarly to Anthropic’s Mythos, deployed within an organization’s IT system to sniff out vulnerabilities.
The security product uses a combination of AI models from Anthropic, OpenAI, and Microsoft to scan, identify, and provide fixes. Its main selling point will be its cost, expected to be far lower than Anthropic Mythos, which has an estimated API cost that is 100 percent higher than Opus and 82 percent higher than GPT, two of the most expensive publicly available AI models.
It plans to reduce costs by routing individual queries to specific AI models based on the task, according to a report by The Information. This is similar to the approach many tech companies are employing to reduce the costs of AI research and development by adopting cheaper, open-source Chinese AI models for certain tasks and the day-to-day operation of AI services.
Microsoft refocuses AI operations on enterprise customers
Microsoft has refocused its AI operations on returning to the leading edge and offering services to enterprise customers. Its AI lead, Mustafa Suleyman, said he wants to have the company launch frontier AI models next year, without having to distill from other AI models. It is also launching hardware devices specifically for AI usage.
The refocus has also sharpened Microsoft’s attacks on its partners turned competitors: OpenAI and Anthropic. CEO Satya Nadella said these AI models were duping customers by using their data to improve their own models, while insinuating that Microsoft had controls in place to ensure it wouldn’t do something similar.
Finding ways to pull AI customers away from these popular AI models and services is a challenge, and it seems Microsoft’s pitch is twofold: we are better at security and governance, and we are cheaper than the competition. Because of Windows and Office, it has a foothold in the enterprise market, even though Anthropic, OpenAI, and Gemini have more popular AI models.
More Microsoft news
- Inside Microsoft’s Real-Time War Against Cybersecurity Threats
- Project Ire: Microsoft Tests AI That Autonomously Detects Malware
- Microsoft Targets ‘Critical AI Talent’ from Meta to Dominate Next AI Breakthroughs
- Windows 10 Support Ends Soon, Though Extended Security Updates Offers Are Available
The race to export AI cybersecurity out of the US
Microsoft is unlikely to face the same pressure and concerns from the US government about launching a security product, although it may have to go through the vetting process OpenAI was subject to with the launch of GPT-5.6. This could give it an edge on Anthropic when it comes to getting a cybersecurity tool to Europe and other regions.
Anthropic had its Fable 5 AI model, which was trained similarly to Mythos but with additional guardrails, banned from export outside the US. This caused Anthropic to shut down access to the AI model while it worked with the US government to verify it was safe to use. Microsoft also restricted access to Fable for Azure customers.
Even with all the obstacles Anthropic has faced in releasing the model, customers still appear set on using it. Several US agencies continue to use Mythos despite the US government designating Anthropic a supply risk, and both governments and financial institutions in Europe have been courting the US government for access to it.
For businesses, the arrival of more security products over the next few months could lower the cost of closing IT vulnerabilities, a growing priority as bad actors gain access to the same advanced tools.
Related reading: The need for faster vulnerability detection is already clear, as a recent critical Zoom flaw exposed Windows users to potential data theft and system compromise before a patch was released.