Good morning, I'm your AI Brief anchor. Here's what's happening in AI today, Saturday, May 9, 2026.
Adept AI Suffers Critical Security Breach
Our top story this morning: a major security crisis at Adept AI. The applied AI company has disclosed a severe breach, resulting in the theft of proprietary model weights and sensitive customer data.
The attackers made off with the core weights for Adept’s upcoming multimodal agent, "Foresight-2." For an AI company, model weights are the crown jewels—the result of years of research and millions of dollars in computing costs. Their exposure represents a catastrophic intellectual property loss.
In addition to the model, an undisclosed amount of customer data was also exfiltrated. Adept says it has notified affected customers and is working with law enforcement and cybersecurity firms to investigate the incident. The breach sends a chilling message to the entire industry, highlighting the immense value of AI models as a target for sophisticated attackers and raising serious questions about the security posture of leading AI labs.
US Government Unveils Landmark 'AI TRUST Act'
The incident at Adept is likely to accelerate calls for stronger oversight, and right on cue, Washington is responding. The United States government has officially announced the "AI Transparency, Responsibility, and Usability in Society Act," or "AI TRUST Act."
This landmark legislation aims to regulate the security and safety of high-risk AI systems. If passed, the act would mandate rigorous third-party security audits and comprehensive red teaming for any AI model deemed critical to national security or public safety. The bill marks a significant shift away from industry self-regulation towards direct government oversight. Proponents argue it’s a necessary step to protect the public from the potential harms of powerful AI, while some industry groups are expressing concerns about the potential to stifle innovation. The AI TRUST Act is now headed to committee, where it is expected to be the subject of intense debate.
OpenAI Pushes Boundaries with Voice and Cybersecurity Models
Moving on, OpenAI is tackling the AI race on two fronts this week, with major announcements in both capabilities and security.
First, the company has launched a groundbreaking update to its API, unlocking realtime voice reasoning. The new models can transcribe, translate, and even reason on streaming audio with near-instantaneous speed. This paves the way for truly natural, fluid conversational AI applications, from real-time assistants to seamless live translation services.
At the same time, OpenAI is working to secure the very systems it helps create. The company has released GPT-5.5 and a specialized version, GPT-5.5-Cyber. This new model is being provided to vetted security experts through its Trusted Access program, giving them a powerful new ally to accelerate vulnerability research and fortify defenses against emerging AI-powered threats. The dual releases show the industry’s central tension: a relentless push for more powerful AI, paired with an urgent race to build the tools to keep it safe.