Microsoft has introduced Project Perception, an agentic security system designed to defend against AI-powered cyber attacks. The system is scheduled for public preview on 3 August.
In its announcement, the software firm says AI-powered attackers operate at machine speed, increasing the scale, complexity and efficiency of cyber threats. It adds that AI-powered threats are eroding the barriers that once protected smaller organisations, and that because AI is used to automate reconnaissance, identify vulnerabilities and scale campaigns at unprecedented speed and lower cost, organisations of all sizes are becoming viable targets.
“Historically, attackers focused their efforts on larger economies and high-value targets, but AI is changing the economics of cyber crime,” says Kerissa Varma, chief security advisor for Africa at Microsoft. “The cost of launching attacks is falling while their reach is expanding, enabling threat actors to target more organisations simultaneously. As a result, businesses across Africa, including those that may previously have fallen below an attacker’s radar, need security capabilities that can detect, analyse and respond to threats around the clock.”
This shift is particularly significant for organisations that have historically operated with limited security resources or less mature security controls.
“Security is an always-on mission, and organisations need protection that is not only highly effective, but also continuously available and cost-effective at scale. Delivering this requires more than access to a single advanced AI model – it means applying the right model to the right task,” says Varma.
Following what is reported to be one of the first publicly disclosed examples of an autonomous AI agent to have successfully compromised production infrastructure, the Cloud Security Alliance has urged organisations to review strategies to defend against AI-driven attacks.
There is also concern over unintentional rogue behaviour, with both OpenAI and Claude developer Anthropic reporting incidents involving advanced AI models that went rogue.
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.
“I believe companies should keep moving quickly with AI adoption. Those that learn how to deploy agents safely will have a meaningful competitive advantage over those that wait. Moving quickly requires a responsible model provider and controls that extend across every system the agent touches. Trusting an AI agent means preparing for paths beyond the one its designers intended, with controls that continue to hold when the agent finds one.”
Defence platform
Project Perception combines signals, context, models and specialised agents into a continuously learning defence platform. Microsoft says the platform uses a multi-model architecture that combines specialised and frontier cyber security models to balance performance and cost.
The new system uses three classes of specialised agents: red team agents, which identify potential attack paths before exploitation; blue team agents, which investigate threats, assess context and determine risk; and green team agents, which take corrective action and strengthen security controls.
South Africa targeted
South African organisations were targeted with an average of 2 065 cyber attacks per week in June 2026, according to research by Check Point Software Technologies. The cyber security company said while SA recorded fewer attacks than Angola, Nigeria and Kenya, the country remains among Africa’s most targeted markets.
According to Microsoft, SA continues to face increasingly sophisticated cyber attacks while organisations must deal with a shortage of cyber security skills. The software company cited the CSIR’s 2024 national cyber security survey, which found that 63% of cyber security roles in SA are partially or fully unfilled. The shortage is particularly challenging outside traditional business hours, when security teams often have reduced capacity.
“The introduction of the first Red/Blue/Green multi-agent autonomous workflow in Microsoft Defender marks an important step forward, enabling organisations to triage alerts, investigate incidents, map attack paths and prioritise remediation across the security stack,” says Varma.
Microsoft says autonomous agents can help organisations investigate threats, prioritise responses and strengthen security operations around the clock. The company adds that technology must be supported by workforce training, security-focused processes and organisational readiness.
Project Perception is built on Microsoft’s security data and intelligence across identities, endpoints, applications, data, clouds and AI systems. The company says its multi-model strategy aims to select the most suitable models for different security tasks while continuing development of its own specialised models.
Share
