Microsoft’s Project Perception Targets AI Security

Microsoft is reportedly preparing Project Perception, an enterprise security product designed to operate inside an organization’s IT system and identify software vulnerabilities. The tool would resemble Anthropic’s Mythos, scanning systems, finding weaknesses, and suggesting fixes.

According to The Information, Project Perception would combine models from Anthropic, OpenAI, and Microsoft, routing each query to the model best suited to the task. That approach could make it substantially cheaper than Mythos, whose estimated API cost is 100% higher than Opus and 82% higher than GPT—two of the most expensive publicly available AI models.

Microsoft’s enterprise AI push

Cost is only part of Microsoft’s pitch. The company has refocused its AI operations on enterprise customers and on returning to the leading edge of model development. Microsoft AI chief Mustafa Suleyman has said the company wants to launch frontier AI models next year without distilling them from other models. Microsoft is also developing hardware specifically for AI use.

The shift has intensified Microsoft’s criticism of former partners and current competitors OpenAI and Anthropic. CEO Satya Nadella has said those companies' models dupe customers by using their data to improve their own systems, while suggesting Microsoft has controls to prevent similar behavior.

Microsoft also has a built-in enterprise advantage through Windows and Office, even though Anthropic, OpenAI, and Google’s Gemini have more popular AI models. Its broader strategy appears to be to compete on security, governance, and price.

Recommended reading

11-byte payload can exhaust OpenSSL server memory

Export restrictions could shape the market

Microsoft may face less pressure from the US government than Anthropic did when releasing a cybersecurity product, although it could still undergo a review similar to the vetting OpenAI faced with GPT-5.6. That could help Microsoft bring Project Perception to Europe and other regions sooner.

Anthropic’s Fable 5, trained similarly to Mythos but with additional guardrails, was banned from export outside the US. Anthropic then shut down access while it worked with the US government to verify the model’s safety. Microsoft also restricted Fable access for Azure customers.

Despite those obstacles, demand for Mythos remains strong. Several US agencies continue using it even after the US government designated Anthropic a supply risk, while European governments and financial institutions have sought access through the US government.

For businesses, more security products arriving over the next few months could reduce the cost of addressing IT vulnerabilities—a growing concern as attackers gain access to the same advanced tools. A recent critical Zoom flaw, for example, exposed Windows users to potential data theft and system compromise before a patch was available.