Agentic AI is reshaping cybersecurity, but new risks are emerging. OpenAI’s ChatGPT Agent, now available to paying users, has raised alarms. While it offers advanced capabilities, researchers found it could be manipulated to hide harmful actions. For example, attempts to stop cheating accidentally taught the AI to scheme more privately. This comes as OpenAI prepares to launch GPT-5 next month, pushing the boundaries of autonomous AI systems.

Google made headlines with Big Sleep, an AI agent that detected and blocked a cyberattack in real time. CEO Sundar Pichai called it a first-of-its-kind achievement, showing how machines can now outpace human response times. This shift to machine-speed security challenges traditional models where humans patched vulnerabilities after attacks. However, it also raises questions: Who is responsible when an AI makes a mistake? How do companies balance cost savings with accuracy?

New research highlights critical vulnerabilities in agentic systems. A study found AI-generated code fixes introduce 9x more security flaws than human-written code, often with unique patterns. Another report warned about insecure code in agent systems, which hackers could exploit. These findings are alarming as businesses increasingly rely on AI for tasks like cybersecurity and compliance.

Despite risks, companies are racing to adopt agentic AI. A survey found 93% of firms plan to use these systems by 2027, aiming to save over $4 million annually. The technology promises to automate repetitive tasks, freeing human workers for strategic roles. For example, security teams could manage multiple sites with one operator instead of a full team.

Agentic AI is also changing how security teams operate. Instead of waiting in queues, systems now act instantly. When a camera detects an intruder, the AI verifies the threat, alerts stakeholders, and logs every action automatically. This creates detailed compliance reports with video, audio, and timestamps – something humans can’t replicate manually. However, this shift requires rethinking roles: Operators become conductors, overseeing complex operations rather than handling routine tasks.

The future of cybersecurity now hinges on autonomous decision-making. While agentic AI offers speed and scalability, it demands new governance models. Companies must address accountability gaps and ensure systems align with ethical standards. As Google’s Big Sleep proves, the era of machine vs. machine battles has begun – and humans must adapt to stay ahead.