OpenAI moves deeper into AI security testing

Security testing moves inside the AI platform

OpenAI just bought Promptfoo.

Promptfoo builds tools that red team and evaluate LLM apps, and the company says more than 25 percent of Fortune 500 firms already use its testing stack.
OpenAI plans to fold the tech into Frontier, its platform for building and operating AI coworkers inside enterprise workflows.

Security is becoming infrastructure. When agents start touching internal data, tools, and approvals, companies need repeatable tests for jailbreaks, prompt injection, and quiet data leaks before anything ships.

The announcement landed in the late afternoon, the kind of quiet post engineers read while CI logs tick across a screen.
This deal shows where enterprise AI platforms are focusing next.

This was predictable. What matters is how quickly everyone else follows.

Microsoft wants every AI agent under supervision

Microsoft thinks companies are about to lose track of their AI agents.

The company announced Agent 365, a system that tracks, governs, and secures AI agents across an organization, alongside a new bundle called Microsoft 365 E7.
Agent 365 becomes generally available May 1 and costs $15 per user per month, while the full E7 suite bundles Copilot and security tools for $99 per user monthly.

This reads like Microsoft preparing for agent sprawl inside big companies.
Once teams start spinning up assistants that can read files, call tools, and act on behalf of employees, someone has to track what exists and what those agents are actually doing.

The dashboard idea is simple: every agent shows up in one registry with an identity, activity logs, and risk signals from Defender, Entra, and Purview.
You can picture an admin screen glowing in an open office at 8:30 a.m., showing dozens of agents running across email, documents, and internal apps.

If companies need a control tower for AI agents already, how crowded does that screen look in two years?

Robotics startups are pairing up with chip companies

Robotics companies are starting to pick their chip partners early.

German startup Neura Robotics announced a partnership with Qualcomm to build the core computing stack for humanoid and industrial robots.
Neura plans to run its robots on Qualcomm’s Dragonwing IQ10 processors and test them inside its Neuraverse simulation platform, which launched in 2025.

This looks like the robotics version of the smartphone supply chain forming.
Instead of buying chips off the shelf later, robotics companies are working directly with chip makers so hardware, software, and training systems develop together.

The pattern is showing up elsewhere.
Boston Dynamics teamed up with Google DeepMind in January to speed up work on the Atlas humanoid robot, and deals like this are likely to multiply as Nvidia, Qualcomm, and others push into physical AI.

If robots become a real market, the winners may be decided long before the machines leave the factory floor.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.