Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

One of the biggest security problems is speed, not a lack of tools. Constant pressure, tired judgement and endless cognitive load make people more likely to miss phishing, mishandle data and make poor decisions, so slowing down becomes a practical security control.

AI systems treat almost everything as a possible input channel, from PDFs and PowerPoints to text, music and more. Without a real separation between control plane instructions and data plane content, the attack surface keeps growing faster than most security programmes can contain.

People are installing powerful AI agents on everyday laptops without realising those tools can access files, emails and operating system functions. Once prompt injected, that agent can behave like a malicious version of its user, which turns convenience into a direct path for deletion, exfiltration and loss of control.

A threat actor reportedly used a prompt injected AI model under the cover of a fake bug bounty story to help compromise government systems and extract large volumes of data. The case shows how jailbroken AI can become an active force multiplier for offensive operations, not merely a productivity tool.

AI adoption is happening faster than any technology cycle in history. Information security and risk management are being sacrificed for speed and every single technology revolution has followed the same pattern. In this episode of Razorwire Raw, Jim Rees draws on decades of experience through the internet boom, virtualisation revolution and cloud computing adoption to explain what's actually happening with AI right now. Each cycle has been faster than the last, and each time, security gets left behind.

Many organisations are pushing AI adoption at full speed while treating security as something to patch in later. The result is a market full of rapidly deployed models, third party tools and jailbreakable systems where capability gets rewarded first and control gets ignored until damage appears.

Prompt injection works because language models struggle to tell the difference between trusted instructions and untrusted user content. Unlike SQL injection or cross site scripting, there is no clean deterministic defence, which leaves code, libraries and AI workflows open to manipulation at multiple points.

What happens when your organisation adopts AI faster than your security strategy can keep up? Welcome to Razorwire, the podcast where we share our take on the world of cybersecurity with direct, practical advice for professionals and business owners alike. I’m Jim and in this episode, I’m joined by Martin Voelk, penetration tester and AI red teamer, and Jonathan Care, lead analyst covering the intersection of AI, cybersecurity and identity.

AI agents with browser access can be manipulated into downloading and running malware disguised as a harmless support tool. It is social engineering for machines, where prompt injection turns an innocent browsing task into a full system compromise without any classic CVE in sight.

Regular pen tests and occasional vulnerability scans no longer match the speed of AI driven development and daily changes in modern systems. The real risk now sits in fast moving exposures, chained attacks and agentic behaviour that traditional CVE focused thinking was never built to handle.