AI News: September 3, 2026

1. DOJ Files in Support of Fair Use for AI Training

U.S. Department of Justice. The DOJ sided with the AI companies in the New York Times case against OpenAI and Microsoft, arguing that training on copyrighted material is fair use and that model outputs often, if not always, lack substantial similarity to the originals. The filing contradicts the U.S. Copyright Office’s earlier report, which rejected a blanket fair-use theory for training at scale. The case is widely treated as the bellwether for whether labs can train on copyrighted works without licensing. Source

2. Astra’s Recurrent Depth Reasoning Draws Safety Objections

OpenAI. Astra uses recurrent depth, also called opaque recurrence, processing queries in loops rather than sequentially and leaving fewer legible traces than chain-of-thought. Buck Shlegeris warned that pushing the technique further gives OpenAI the option to massively increase recurrence and destroy CoT monitorability, and Ryan Greenblatt raised the prospect of a model reasoning almost entirely in latent space. Chief scientist Jakub Pachocki responded that OpenAI has worked to preserve chain-of-thought monitoring since its first reasoning models and that Astra’s use of the technique is limited. Source

3. Court May Force Disclosure of Federal AI Safety Testing Rules

U.S. government. A legal challenge could compel the Trump administration to disclose the currently undisclosed rules federal agencies use when running AI safety tests. The outcome would determine how much visibility the public gets into the criteria behind government model evaluations. Source

4. Wonderful Raises $550M at a $5B Valuation

Wonderful. The enterprise AI software company closed a $550 million Series C led by Insight Partners at a $5 billion post-money valuation, more than doubling its valuation in under six months. Source

5. HiddenLayer Raises $100M to Secure AI Models and Agents

HiddenLayer. The Series B was led by Delta-v Capital with participation from Ten Eleven Ventures, Morgan Stanley, Microsoft’s M12, and Booz Allen Hamilton. HiddenLayer sells tooling against prompt injection, agent manipulation, malicious tool use, and model-level attacks, and says ARR grew more than 10x over the past year into the tens of millions, with over 90% of that growth from customers acquired in the last year. Its customer list includes financial services firms, large tech companies, the Department of Defense, and intelligence agencies. Source

6. Palo Alto Networks Paid $500M for Console

Palo Alto Networks. Sources put the price of the Thrive-backed Console acquisition at $500 million, continuing the security vendor’s run of consolidation as enterprises look for a single vendor across AI and cloud security. Source

7. Adobe Acquires Market Intelligence Startup Rilo

Adobe. Adobe acquired Rilo, an India-based market intelligence startup, adding its capabilities to Adobe’s analytics and marketing stack. Terms were not disclosed. Source

8. Amazon’s Shopping Assistant Adds Scam Message Detection

Amazon. Amazon’s shopping AI can now assess whether a message a shopper received is a scam, extending the assistant from product discovery into fraud screening for the messages buyers get around a purchase. Source

9. India’s Richest Man Targets Aging PCs as AI Hardware

Reliance. The plan is to convert aging computers into AI-ready PCs rather than replace them, aiming at the installed base of older machines in India as a cheaper on-ramp to local AI workloads. Source

10. Trump Frames Data Center Protests as Helping China

U.S. government. Trump argued that local opposition to AI data center construction plays into China’s hands, tying siting fights over power and water to the competitive framing that has driven U.S. AI infrastructure policy. Source