AI News Roundup — August 3, 2026
The White House pulled OpenAI, Anthropic, Google, and Meta into talks on voluntary AI safety testing as China's model race intensified with Alibaba's giant Qwen3.8-Max and ultra-cheap DeepSeek V4-Flash, while OpenAI published a pointed public rebuttal to Apple's trade-secret lawsuit.
White House Convenes Top AI Labs Over Voluntary Safety-Testing Framework
The White House invited OpenAI, Anthropic, Google, Meta, and other leading AI developers to discussions on a voluntary framework for testing advanced models before deployment, with an initial focus on cybersecurity evaluations. The move is one of the administration's first concrete steps toward common safety-testing norms without imposing comprehensive federal regulation, though it leaves open whether voluntary commitments will prove sufficient.
Alibaba Unveils Qwen3.8-Max, Its Largest Model Yet
Alibaba launched Qwen3.8-Max, its most capable model to date at roughly 2.4 trillion total parameters with about 95 billion active per token under a mixture-of-experts design, targeting coding, long-horizon agentic tasks, and multimodal reasoning. Alibaba said open weights will follow, alongside a smaller Qwen3.8-27B variant. The launch is widely read as Alibaba's answer to Moonshot AI's 2.8-trillion-parameter Kimi K3, reinforcing China's strategy of competing via massive MoE architectures and increasingly open weights.
Reuters · Latent Space · Global Times
DeepSeek V4-Flash Touted as World's Cheapest Frontier-Class Model to Run
Coverage highlighted DeepSeek V4-Flash, which became available the previous Friday, with a benchmark-cost analysis claiming it can be more than 100 times cheaper to run than Anthropic's Claude Fable 5 on the cited test. The report also noted DeepSeek is reportedly preparing for a possible IPO, though no transaction has been confirmed. The cost comparison rests on one research firm's methodology and shouldn't be read as a universal real-world cost measure.
OpenAI Fires Back at Apple in Trade-Secret Lawsuit: "Apple Is Getting This Wrong"
OpenAI published a detailed public rebuttal to Apple's trade-secret suit (Apple Inc. v. Liu) after Apple sought a preliminary injunction on August 3. OpenAI says Apple's lawyers emailed the wrong person before filing suit, disputes claims about former Apple engineer Chang Liu's file access, and defends new Chief Hardware Officer Tang Tan, publishing redacted iMessage exchanges it says show Apple employees — not Liu alone — sought access to his old accounts after he left. OpenAI says it holds no Apple trade secrets and the injunction is unnecessary; Apple's allegations remain unresolved by any court.
U.K. Signals Mandatory AI Regulation Remains on the Table
U.K. AI Minister Kanishka Narayan said Britain would consider regulating advanced AI models if its current voluntary pre-deployment testing regime fails to adequately protect the public — illustrating the broader international pattern of governments favoring voluntary testing for now while keeping mandatory rules in reserve.
EU AI Act Enforcement Powers Expand, Raising Stakes for OpenAI and Anthropic
Reports detailed newly expanded EU AI Act enforcement powers that could affect general-purpose and frontier-model providers, with OpenAI and Anthropic named among companies facing greater regulatory exposure in Europe going forward.
Big Tech's Anthropic and OpenAI Stakes Complicate Corporate Earnings
Corporate investment stakes in OpenAI and Anthropic held by major tech companies were reported to be distorting how those companies present quarterly earnings, underscoring how deeply intertwined Big Tech balance sheets have become with the two leading AI labs' valuations.