“Many small businesses now spend hours each week having employees proofread artificial intelligence,” Ram Varadarajan, CEO at Acalvio, told researchers — a sentence that captures both the scale of adoption and the fragility of oversight uncovered in new findings.
Pax8 report: adoption is rapid, oversight is lagging
The Pax8 report at the center of the findings shows small and mid-sized businesses (SMBs) that engage with AI are moving quickly from experiment to production: two-thirds of AI-engaged SMBs have built solutions that are now in use. That same report also flags control gaps: 35% of these organizations have discovered unapproved workflows, tools, or agents running in their operations, and more than half of decision-makers are spending hours each week checking AI's outputs.
Shadow agents: from chatbot experiments to systems with credentials
Aviv Nahum, Co‑founder and CEO at Above Security, highlights a qualitative shift in risk. “An employee experimenting with a chatbot is one level of risk. An employee building an agent that has credentials, accesses company data, connects to finance or CRM systems, and performs multi-step work autonomously is something very different,” he said. When those agents operate inside core systems, Nahum warned, “the organization has effectively created a new insider, often without a clear owner or a security team even knowing it exists.”
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scramblePractical controls recommended by security practitioners
Security practitioners quoted in the report emphasize behavior-based detection over simple output review. Varadarajan recommends seeding systems “with decoy files and credentials that no legitimate employee or program would have reason to open, so that any contact with them is an immediate warning.” He also advises limiting what an agent can reach before deployment, setting “AI-baiting traps,” and asking technology providers “about what’s monitoring the environment while everyone’s busy watching the chatbot.”
Why policies alone won’t solve it
Nahum cautions that written policies are necessary but insufficient. “Policies alone are not enough. A company can have an AI policy and still have no idea what an internally built agent is doing after deployment,” he said. The core need, according to his assessment, is “continuous visibility into behavior and data movement, especially as AI shifts from generating content to taking actions on behalf of employees.”
How technologists and security teams, decision-makers, and employees are responding
- Technologists and security teams: Expect to prioritize behavior-based monitoring and controls that detect unauthorized access patterns rather than relying solely on output verification. Varadarajan’s suggestion to plant decoy files and credentials points to one concrete visibility technique these teams may adopt.
- Small-business decision-makers and procurement leaders: Face pressure to deploy AI quickly — the report states 66% of SMBs believe their business will not be competitive in the next three years without AI — while also needing to demand better monitoring from technology providers and to assign ownership for internally built agents.
- Employees and builders of internal tools: Are both the source of innovation and the vector of risk. Nahum’s framing shows that an employee-built agent with credentials can function as an uncharted “insider,” creating operational and security exposure if nobody owns or watches its behavior.
The Pax8 findings present a clear tension: the companies getting the most value from AI, the report suggests, are deploying it more broadly, yet that broad deployment converts shadow experimentation into a business-process and security problem. The remedies offered — limit reach before deployment, plant decoys, insist on continuous visibility, and assign ownership for agents and workflows — are all concrete steps rooted in the report’s interviews and analysis. The next practical question the report leaves for small-business leaders and their security teams is simple and urgent: if agents can act like insiders, who will be responsible for watching them and holding them accountable?
