Enterprises Deployed AI Agents Without the Safety Nets. Now They're Paying to Retrofit.

Key points

• 86% of enterprises running their own AI chips report those chips operate at 50% capacity or less, based on a June 2025 survey of 573 technical leaders.
• 54% of companies experienced an AI agent security incident or near-miss in the past 12 months.
• 71% of enterprises say a quarter or fewer of their deployed "agents" can actually complete a multi-step task without a human driving each step.
• 34% already let AI agents push changes directly to production systems with no human review, and another 33% are engineering toward that same setup.
• 69% of companies allow multiple agents to share a single login credential, and those companies suffered security incidents at nearly twice the rate of companies that don't.

Enterprises knew their AI safety controls were not ready. They deployed anyway.

That finding comes from VentureBeat Research's June survey of 573 technical leaders at firms employing 100 or more people. Companies sprinted to ship AI agents (software that handles multi-step tasks on its own, rather than answering a single question) and are now scrambling to bolt on the guardrails they skipped. Roughly six in ten plan to switch or add vendors across key control layers within 12 months.

The bill is real. Fifty-four percent reported an agent security incident, or a near-miss caught before real damage, in the past year.

What does this mean for ordinary workers and customers?

The AI making decisions at companies you work for or buy from may be operating with fewer checks than most people assume. Half of surveyed enterprises shipped an agent that cleared their own internal tests, then caused a customer-facing failure in the real world. A quarter watched that happen more than once.

The hardware story is equally striking. Eighty-six percent of enterprises running their own GPUs (specialised chips that handle AI's heavy computation) say those chips run at half capacity or less. Firms spent heavily on AI infrastructure, and most of it sits underused. Yet 45% plan to evaluate a specialist AI cloud provider within 12 months, and roughly one in three is weighing chips not made by Nvidia. AI2Day has tracked GPU utilisation concerns across three stories since 14 July 2026, including our earlier finding that nearly a third of senior leaders can't track what AI is costing them at all.

The agent label itself is largely marketing. Only one in ten enterprises says true autonomous agents make up the majority of what they run. Gartner calls the broader habit of labelling basic chatbots as agents "agentwashing," a term we first covered on 14 July 2026.

Security is the sharpest near-term risk. Sixty-nine percent of companies let multiple agents share one login credential. Those organisations suffered incidents at a 63.5% rate, against 40.9% at firms where every agent holds its own separate, limited-access identity.

Wrong answers add a quieter liability. Fifty-seven percent of enterprises traced a confident, incorrect agent output back to missing or stale internal business data.

Should you trust automated agent decisions?

Two-thirds of firms are moving that way regardless. Only 5% say they fully trust the automated checks that would replace human review. Before removing a person from any approval step, the data says to test your automated checks against real production failures, not just internal benchmarks.

Measure what you have before buying more. Check chip utilisation. Give every agent its own login. That's not a complicated fix, it's just the work that got skipped the first time around.