OpenAI is rolling out a more capable version of its GPT-5.5-Cyber model, built specifically for vetted cybersecurity teams and researchers.
The model can scan large codebases, find weaknesses, test vulnerabilities, and help create software fixes much faster than traditional methods.
It highlights one of the biggest debates happening around advanced AI.
These models could become some of the best tools ever created for defending software systems, while also creating new risks if the same capabilities become widely available.
OpenAI’s solution is controlled access: trusted researchers and companies get the most advanced tools, while public models remain more restricted.
As AI gets better at finding and fixing security flaws, access to the strongest models could become a major advantage for the organizations allowed to use them.
Supported by Unblocked
Not long ago, you were the context engine for your coding agents— every good session ran on what you remembered to paste in. Then teams moved to rules, skills, and MCP’s, and most quietly stalled there. Output is still inconsistent despite the climbing token spend.
Join live June 24 (FREE) to learn:
Which level of context maturity your team is at
Why more MCPs and rules eventually stops helping (and what to do instead)
What it takes to build a context layer that lets humans step out of the loop
Nvidia is expanding deeper into robotics with Halos, a new safety system designed to help humanoid robots work alongside humans.
The system combines AI software, chips, sensors, and safety tools built from Nvidia’s experience with self-driving cars. The goal is to help robots understand their surroundings, avoid dangerous actions, and operate safely in factories, warehouses, and other real-world environments.
Humanoid robots are still early, but companies are already preparing for a much bigger shift. Barclays estimates the market could reach $200 billion by 2035 as robots move from demos into actual workplaces.
Nvidia’s strategy looks very similar to its AI playbook.
The company became the backbone of the AI boom by powering the models everyone else built. Now it wants to do the same thing for the machines bringing AI into the physical world.
Anthropic has hired John Jumper, the Nobel Prize-winning researcher behind one of Google DeepMind’s biggest scientific breakthroughs.
Jumper helped develop AI systems that can predict protein structures, work that earned him and DeepMind CEO Demis Hassabis the Nobel Prize in Chemistry in 2024.
His move comes shortly after Google lost another major AI researcher, Noam Shazeer, to OpenAI.
The biggest AI labs are fighting for researchers who can push the technology into entirely new fields like biology, medicine, and scientific discovery.
As OpenAI, Anthropic, and Google race toward more powerful systems, the people behind the breakthroughs are becoming some of the most valuable assets in tech.
In other developments
SpaceX has signed a $6.3B data center deal with AI startup Reflection AI, expanding its role in the AI infrastructure race as demand for compute power continues to surge.
Apple’s iOS 27 is bringing AI beyond Siri, adding everyday features like automatic bill splitting, password fixes, smarter Messages suggestions, call context, AI-built Shortcuts, organized Safari tabs, and cleaner smart home alerts.
Nvidia says its next-gen AI systems could solve data centers’ water problem, using advanced liquid cooling that reduces the need for water-heavy chilling equipment as AI infrastructure rapidly expands.