AI News for June 11, 2026 — 7min.ai

[RESEARCH] Google DeepMind releases DiffusionGemma, generating text 4x faster than autoregressive models: DeepMind shipped DiffusionGemma under Apache 2.0, a 26B-parameter Mixture-of-Experts model with 3.8B active parameters that produces 256 tokens in parallel using diffusion instead of next-token prediction.
[OPINION] Anthropic CEO calls for FAA-style AI regulation, commits $350M to labor displacement: Dario Amodei published "Policy on the AI Exponential," calling for federal authority to block frontier model deployments deemed catastrophically risky, modeled on FAA pre-launch review.
[NEWS] OpenAI in talks to lease 10GW Ohio data center with Nvidia guaranteeing $500B in payments: OpenAI is negotiating a 20-year lease on a 10-gigawatt data center in southern Ohio, with Nvidia acting as guarantor for the lease and project financing.
[NEWS] Neura Robotics raises $1.4B from Nvidia, Amazon, and Tether to ship humanoid robot: German startup Neura Robotics closed roughly $1.4B in a round backed by Tether, Nvidia, and Amazon to develop and ship its AI-powered humanoid robot.
[NEWS] OpenAI exposes China-linked influence operation pushing US data center backlash: OpenAI banned a set of likely China-based ChatGPT users running a campaign it dubbed the "Data Center Bandwagon" Campaign, which posed as Americans on X and Facebook to amplify opposition to US AI data centers.
[RESEARCH] GPT-5.5 beats Claude Fable 5 on UC Berkeley's brutal Agents' Last Exam benchmark: UC Berkeley's RDI center launched Agents' Last Exam, a benchmark of multi-hour, economically valuable professional workflows graded by 300+ domain experts.
[RESEARCH] Sapient trains 1B foundation model from scratch for $1,500 with hierarchical recurrent architecture: Sapient researchers released HRM-Text, a 1B-parameter model trained from scratch using a Hierarchical Recurrent Model instead of a standard Transformer.
[RESEARCH] Anthropic study: AI cuts patch-to-exploit time from weeks to a single afternoon: Anthropic's security team measured how fast Claude models can reverse-engineer N-day vulnerabilities from Firefox and Windows security patches.
[NEWS] Altman tells staff OpenAI IPO is 12 months out, sets employee tender at $687.69/share: Sam Altman told OpenAI employees on Slack that the IPO is expected "within the next year," a longer horizon than the fall window cited in this week's confidential S-1 announcement.
[NEWS] Microsoft restricts Claude Fable 5 internally over Anthropic's 30-day data retention: Microsoft is blocking Claude Fable 5 in the internal GitHub Copilot model picker used by its own employees, even after rolling the model out broadly to GitHub Copilot and Azure Foundry customers.

About 7min.ai · Subscribe to newsletter