What happened
Pretxt, a startup with offices in Aspen and in Dublin, Ireland, has come out of stealth with $5 million in seed funding, SecurityWeek reported. Deciens Capital, Howdy Partners and Service Provider Capital backed the round. The company is launching a product it calls AI Supervision, a runtime tool that oversees AI agents while they operate and corrects their behavior before errors turn into business problems.
According to co-founder and CTO Aaron Costello, customers connect their agents to the platform. Pretxt can then read each agent's instructions, observe how it communicates and map the tools and actions available to it, such as sending email or updating a database. Customers then set out what an agent must never do or may only do. That guidance becomes the reference Pretxt's own AI uses to judge each action.
Pretxt redirects an agent that oversteps instead of halting it. Costello gave the example of a medical practice whose booking assistant must not offer medical advice. Pretxt inspects each exchange with patients in real time and blocks any advice it detects. It then quietly instructs the agent to keep helping with the appointment, so patients can still book.
Context
Pretxt joins a newly forming group of cybersecurity vendors that try to identify harmful agent actions as they occur. The report traces rogue agent behavior to ambiguous instructions combined with broad agent autonomy. It also notes that ambiguity in language cannot be fully removed, while tightly restricting what agents can do may sacrifice significant business value.
The briefing
The week in tech, in one email.
The stories worth your time, every week. Unsubscribe anytime.
