Pretxt, a newly launched enterprise AI company focused on runtime oversight of AI agents, has stepped out of stealth mode with $5m in seed capital to build out a technology category it calls AI Supervision.
Deciens Capital, Howdy Partners, Service Provider Capital and NSV Wolf Capital backed the round. Pretxt plans to use the capital to bring its supervision layer to enterprises that want AI agents to handle higher-stakes work without drifting from approved procedures.
AI agents no longer simply produce text. They now speak with customers, connect to other systems, handle transactions and run business processes. According to Pretxt, these agents do not always behave as instructed.
They may skip mandatory steps, create their own exceptions, make commitments nobody authorised or ignore sign-off requirements. A support agent offering a refund without approval, or a financial services agent that sidesteps identity checks, can expose a business to operational, financial and reputational damage, and none of this needs a hacker to trigger it.
Pretxt argues that security, governance, guardrails and observability tools each tackle part of the risk, yet none confirms that an agent is completing its task in line with business or regulatory standards.
Some guardrails halt prohibited actions and leave processes stuck awaiting human help, while monitoring tools mainly explain failures after they occur. The company cites figures showing 83% of US consumers believe brands should answer for AI interactions that go wrong.
Pretxt defines AI Supervision as continuous oversight of agent behaviour during execution, judged against an organisation’s own processes, intent and procedures.
Source link







