{
  "id": 3003952,
  "title": "Why most organizations are getting AI security wrong (and why it’s about to catch up with them)",
  "url": "https://urgent.news/2026/08/24/why-most-organizations-are-getting-ai-security-wrong-and-why-its",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-24T10:42:21.000Z",
  "source": {
    "name": "TechRadar",
    "slug": "techradar",
    "url": "https://www.techradar.com/pro/why-most-organizations-are-getting-ai-security-wrong-and-why-its-about-to-catch-up-with-them"
  },
  "original_language": "en",
  "account": "The world of artificial intelligence is moving at a rapid pace, with AI technologies being rapidly adopted, integrated into products, and deployed for real-world use. However, as businesses become more comfortable with AI and discover new ways to utilize it, there is a growing concern that threatens their cybersecurity. While organizations are quick to implement AI, they are often neglecting the crucial aspects of its security. The gap between the rapid adoption of AI and the lack of proper security measures is widening, leaving security teams scrambling to address vulnerabilities.\n\nAI doesn't behave like traditional applications, making it a unique challenge to secure. Instead of following a conventional lifecycle, AI goes from experimentation directly to becoming a critical component of business processes, often with APIs, models, agents, and data sources that weren't originally designed to work together. This creates a more fluid, dynamic tool that behaves differently, making it harder to secure.\n\nTraditional security measures, which are usually applied at edges, monitoring outcomes and analyzing behavior after the fact, are proving ineffective in this new landscape. Instead, AI security needs to focus on the runtime, where decisions are made and where the risks can occur in real time. Prompt injection, where models can be manipulated and sensitive data can leak, is a significant concern. This isn't just about incorrect configurations; it's about the continuous chain of events in AI interactions where things can go wrong unexpectedly.\n\nThe industry is starting to recognize the importance of securing AI at the runtime, but many organizations are still getting it wrong. This is often due to the decisions being made by innovation teams or developers, who implement AI tools behind the scenes. As a result, infrastructure and security decisions follow behind rather than shaping the architecture from the start.\n\nTo truly secure AI, a new approach is needed. It's not about adding standalone security tools but about rethinking control. Security should be placed in the flow of traffic itself, where behavior can be influenced the most and where policy can be enforced. This shift moves AI security from being an isolated concern to being embedded in how systems operate. This change isn't tied to a single vendor but reflects the evolving nature of AI and the need for control to define the next era of AI security.",
  "summary": "Most organizations are securing AI incorrectly, leaving critical runtime vulnerabilities exposed as enterprise adoption accelerates.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}