Thousands Of Unrestricted AI Models Are One Click Away. Are Midmarket IT Teams Ready?

New research raises the question for midmarket IT leaders: What happens when the same technology helping defenders becomes readily available to bad actors?

Thousands of unrestricted AI models are now just a few clicks away, according to new research from cybersecurity vendor ThreatDown, adding a new speed bump in the industry’s rapid adoption of artificial intelligence.

ThreatDown’s 2026 “Cybercrime in the Age of AI” report identified more than 6,600 publicly available AI models on Hugging Face that were labeled with “uncensored,” “unfiltered,” “decensored” or otherwise. The labels indicate the models operate without the safeguards commonly found in commercial AI systems.

Researchers said those models had been downloaded more than 22 million times during a single 30-day period.

The findings come as organizations push to deploy AI across security operations, from threat detection to incident response.

The report also poses the question for midmarket IT leaders: What happens when the same technology helping defenders becomes readily available to bad actors?

According to ThreatDown, many criminal AI services are no longer developing their own underlying models. Instead, they are increasingly opting for the same frontier AI technologies, cloud infrastructure, and commercial services used by legitimate businesses.

The report also found that malicious AI is shifting from running on hosted services toward locally run models that can evade monitoring and restrictions.

Security Experts Warn The Barrier To Entry Is Faling

Security researchers say the significance of unrestricted AI models is not that they introduce new attack techniques, but that they make sophisticated capabilities more accessible and faster to use.

"The availability of these models provides a valuable data point of capabilities that traditionally only nation-states or criminal organizations had access to," said Ben Actis, an AI security researcher at Straiker.

Actis pointed to the recent Hugging Face compromise as an example of how quickly capable attackers can move. "Within just two days, a small actor went from initial breach to significant compromise of Hugging Face's production network," he said. "It would cost $15,000 to $50,000 a month to stand this up. This isn't a significant cost at all to do."

Neil "Grifter" Wyler, vice president of defensive services at Coalfire, argues that unrestricted AI is accelerating existing attack methods rather than creating entirely new ones.

"The surge in unrestricted AI models represent a collapse of enterprise security timelines rather than a fundamental invention of new attack techniques," Wyler said. "While the playbook remains familiar, AI acts as an accelerant, compressing a kill chain that once took days or weeks into mere minutes."

For midmarket organizations, Wyler said the development could fundamentally alter risk calculations because many security programs were designed around threats operating at human speed.

"While there are increasing government calls to regulate or slow the release of frontier models from labs like OpenAI or Anthropic, unregulated and uncontrollable open-source models are closing the capability gap to those frontier models in months, sometimes even weeks," he said.

Security Fundamentals Still Matter

Despite red flags about unrestricted AI, experts caution against solely focusing on the models themselves.

"Attackers—even those using AI—still prioritize the path of least resistance," Wyler said. "Organizations should view this moment as a starting gun to close visibility gaps and fix foundational architectural flaws, as you cannot defend against an autonomous adversary if you haven't solved the security fundamentals."

While the availability of thousands of unrestricted AI models may not change the basic tenets of cybersecurity, for the midmarket, the real challenges may be how best to adapt security operations to a cybersecurity battle zone where both defenders and attackers increasingly have access to the same powerful AI tools—and where the speed of attacks may be accelerating faster than many organizations are prepared to handle.