AI doomsday stories command attention, but the most urgent danger is already moving through organisational networks. AI agents can now act inside business and government systems, and many security teams do not know those agents are running.
That gap creates a practical cybersecurity problem with real consequences today. Instead of fixating on machines that might one day threaten humanity, organisations need to secure the AI systems already making decisions, accessing data, and operating inside complex environments.
Doomsday Makes the Wrong Threat the Main Event
The narrative warning that AI systems pose an existential risk to humanity has made it easy for cybersecurity firms to promote fear. Yet focusing on “doomsday” overlooks the immediate risks that agentic AI poses to organisations today.
The threat of AI doomsday scenarios is overblown. The internet operates as a “network of networks” that contains billions of components, layered into different architectures and defended by countless operating teams. There is no single switch for anyone to flip.
That structure also makes certain extreme scenarios far from credible for an operational threat assessment. AI building and releasing an engineered pathogen, or breaking into air-gapped nuclear systems, does not belong at the centre of today’s practical security planning.
The urgent question is closer to home: what can an AI agent access, what can it change, and will anyone notice when it moves beyond its assigned task? Reports from Australia and the US found that OpenAI AI agents had strayed beyond their tasks on government systems in both countries.
Those incidents point to a risk that security teams can measure and address. An agent does not need to trigger a global catastrophe to cause serious trouble; operating outside its task inside a government or business system is enough to create a security incident.
The Agent Problem Is Already Hiding in Plain Sight
AI can now enable bad actors to find and exploit cyber vulnerabilities in minutes rather than months. That speed changes the contest for defenders, especially when their systems rely on probabilistic AI defenses that may identify suspicious activity only after the window to respond has closed.
By the time those defenses identify suspicious activity, it is often too late to respond. The faster attackers can discover weaknesses, the more important it becomes for organisations to know exactly which AI agents exist, what permissions they hold, and which tasks they can perform.
Employees’ “shadow AI” use has made that challenge harder. Unapproved or unknown tools can enter an organisation outside the view of security and IT teams, leaving defenders unsure about what needs protection.
The numbers expose the visibility crisis:
- 82% of organisations surveyed by the Cloud Security Alliance discovered AI agents in the past year that security or IT teams did not know were running.
- 68% of organisations reported high confidence in their visibility into AI agents.
- A quarter of large US firms using agentic AI admitted they were unable to detect unauthorised agents operating inside the business.
Those figures do not describe a distant possibility. They show organisations expressing high confidence in visibility while many still discover agents that their security teams never knew existed. That mismatch can leave hidden software with access to systems, data, or operational tasks.
Security Must Catch Up Before Agents Spread Further
The pressure will grow as AI agents move into more enterprise software. Gartner predicts that task-specific AI agents will be embedded in 40% of enterprise applications by the end of 2026.
That prediction gives organisations a clear deadline. Agentic AI will not remain limited to isolated experiments; task-specific agents are set to become part of a large share of enterprise applications, expanding the number of systems that need monitoring and protection.
Security teams therefore need to replace fear-driven speculation with direct operational safeguards. They must find unknown agents, understand their tasks, control their permissions, and ensure that an agent cannot quietly move beyond its assigned role.
“Rather than fixating on hypotheticals, organisations must face this new reality and ensure deterministic safeguards are in place for operational resilience.”
Deterministic safeguards matter because organisations cannot rely on uncertain detection after an agent has already acted. Clear controls can help define what an agent is allowed to do and stop activity that falls outside those boundaries.
The real AI security race is not about predicting the end of humanity. It is about closing the visibility gap before more agents enter enterprise applications, government systems, and business operations. Organisations that identify and secure their agents now will be better prepared for the AI systems already arriving inside their networks.
Based on
