Microsoft has released an agent-based security system called Project Perception to combat AI-driven cyber threats. The platform will enter public preview on August 3.
The system addresses attackers who use AI to automate reconnaissance, exploit weaknesses, and expand campaigns rapidly. These capabilities reduce the cost of cybercrime, making smaller organizations vulnerable targets for the first time.
AI changes cybercrime economics
“Attackers once focused on larger economies and high-value targets, but AI is altering the cost structure of cybercrime,” said Kerissa Varma, Microsoft’s chief security advisor for Africa. “Launching attacks now costs less while their reach grows, allowing threat actors to target more organizations at once.”
This change hits businesses with limited security resources hardest. Varma explained that protection must be constant, requiring solutions that are both effective and affordable at scale. Meeting this need, she added, requires more than a single advanced AI model—it demands using the right model for each task.
The need for AI-driven defenses follows reports of autonomous agents breaching production infrastructure. The Cloud Security Alliance has urged organizations to review their strategies against such risks. Concerns also include unintended behavior, with incidents reported by OpenAI and Anthropic, where models acted unpredictably.
Project Perception’s approach
Project Perception combines signals, context, and specialized agents into a learning defense platform. Its architecture blends advanced cybersecurity models with task-specific ones to optimize performance and cost.
Related: OpenAI unveils AI that hunts its own flaws
The system uses three types of agents:
- Red team agents test for potential attack paths before exploitation.
- Blue team agents investigate threats, assess context, and evaluate risk levels.
- Green team agents take corrective actions and strengthen security controls.
The platform leverages Microsoft’s security data and intelligence, covering identities, endpoints, applications, data, clouds, and AI systems. Its strategy deploys the most appropriate models for different tasks while developing specialized models internally.
In June 2026, South African organizations faced an average of 2,065 cyber attacks weekly. Though the country experienced fewer attacks than Angola, Nigeria, and Kenya, it remains one of Africa’s most targeted markets. A CSIR survey found 63% of cybersecurity roles in South Africa are partially or fully vacant, with shortages most severe outside business hours.
The lack of skilled personnel makes automated defenses especially useful. Yet even the best systems need human oversight.
In a blog post, Adam Ely, GM for AI security at Check Point Software Technologies, said advanced AI agents can pursue objectives in unexpected ways, making runtime governance and security controls increasingly important.
The public preview of Project Perception will test whether Microsoft’s approach can meet these challenges. Firms looking to mitigate risks in fast-growing operations may find such tools increasingly necessary.
