Governor Greg Abbott ordered ERCOT and state regulators to audit every new data center before it can connect, with 474GW now queued.
The biggest US market for AI compute just put a screening gate on grid access, resetting where and how fast frontier capacity can be built.
⚡ Quick hits
IndustryOllama said DeepSeek-V4-Flash-0731 is its fastest-growing model ever by token usage and that it is scaling capacity in the US and Europe.X @ollama ↗
SafetyThe UK AI Security Institute published a cybersecurity evaluation of Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6 Sol, which Anthropic amplified.X @AnthropicAI ↗
The nonprofit's first AI Security Leaderboard reports a hundredfold spread in safeguard robustness, with two frontier models yielding none.
⚡ Quick hits
SafetyOpenAI disclosed two incidents that occurred during third-party cyber evaluations of its models, detailing what happened and its remediation steps.OpenAI ↗
AIGCPika launched the Pika API Club, a $10-per-month membership offering discounted access to 100-plus video, image, audio and LLM models through one API.X @pika_labs ↗
AgentsAWS made Web Search generally available on Amazon Bedrock, grounding models in Amazon's web index via one API parameter with zero data egress.AWS Machine Learning Blog ↗
ResearchArtificial Analysis launched an Endpoint Accuracy Index measuring how much of an open-weights model's accuracy each serverless API endpoint actually preserves.X @ArtificialAnlys ↗
Mixture-of-Kittens fuses all MoE communication and compute into one deterministic Blackwell kernel, lifting Cursor's training throughput 1.41x.
⚡ Quick hits
AIGCSpotify said Merlin, representing over 30,000 independent labels, joined Universal in backing its opt-in paid AI remix and covers tool.TechCrunch ↗
AgentsCognition shipped Devin Fusion in Devin Cloud, saying harness and model gains make it 4% more intelligent and 27% cheaper on FrontierCode 1.1.X @cognition ↗
The six-year deal routes Anthropic through a 133MW hydro-powered Norwegian site running Nvidia Vera Rubin systems, operated with Bitdeer.
⚡ Quick hits
AgentsMem0 launched Dream, a background layer that continuously reconciles contradictory memories inside agent projects instead of leaving conflicts to accumulate.X @mem0ai ↗
AIGCTripo AI won Best in Show at SIGGRAPH Real-Time Live for on-stage game-ready 3D asset generation, with the audience creating over 600 models live.X @tripoai ↗
Open SourceDatabricks released Omnigent, an open-source AI meta-harness for composing and governing coding and workflow agents in one place.X @databricks ↗
ModelsLiquid AI released LFM2.5-2.6B for on-device agents, claiming 220 tokens/second on an Apple M5 Max within 2.5GB of memory.Hugging Face ↗
ResearchMicrosoft Research introduced PRISM2, a pathology foundation model trained on 2.3 million whole-slide images and millions of report-derived question-answer pairs.X @MSFTResearch ↗
IndustryOpenAI countered Apple's trade-secret suit with iMessage logs showing Apple staff repeatedly texted a departed engineer for internal information, arguing Apple's own access controls failed.The Decoder ↗
Tencent's new speech model reports about 3% word error rates in Mandarin, English and Cantonese, and is live on Tencent Cloud and in Yuanbao.
⚡ Quick hits
Open SourceReflex open-sourced XY under Apache-2.0, a Rust-backed Python charting library holding ~0.08s renders from 10,000 to 100 million interactive points.MarkTechPost ↗
AIGCRecraft showed V4.1 swapping an object's material while keeping its form intact, pitching editable physical logic rather than a freshly regenerated image.X @recraftai ↗
ResearchA Kansas mathematician refuted OpenAI's claimed disproof of the Connes rigidity conjecture within 24 hours, tracing 37,000 lines of Lean 4 to a group failing preconditions.量子位 ↗
IndustryOpenAI's luxury Hudson Valley creator retreat drew backlash over $2,000-a-night rooms and farm activities, with several influencers deleting their posts afterwards.量子位 ↗
AgentsMarkTechPost detailed a skill-auditing pipeline using NVIDIA SkillSpector and LangGraph to scan agent skills for shell injection, credential harvesting and MCP attacks before deployment.MarkTechPost ↗
FundingOm AI raised several hundred million yuan led by Qianhai Mother Fund for edge-native multimodal perception models, whose Homer AI serves nearly 100,000 visually impaired users.量子位 ↗
AgentsY Combinator open-sourced QM under MIT, a multiplayer agent harness for Slack and web giving each person isolated memory, files, permissions and sandboxes.MarkTechPost ↗
PolicyMiniMax rebutted claims its H3 video model cannot be legally used in certain regions, saying it can be licensed for deployment in the US, EU, UK and South Korea.X @MiniMax_AI ↗
SafetyA study of 414 internet-facing MCP servers found 68 reportable flaws and 91.8% lacking OAuth, plus 687 tools offering unrestricted shell execution.arXiv ↗
Wan launched a Real-time feature suite whose first capability turns a live phone camera feed into stylised imagery, starting with anime looks.
⚡ Quick hits
AIGCPixVerse showed a rooftop-to-office scene built in PixVerse Canvas, saying the physics holds up down to a tote bag swinging correctly through the shot.X @PixVerse_ ↗
IndustrySakana AI became a full member of Japan's AI Robot Association this month, collaborating on world models and physical AI research under its RSI Lab.X @SakanaAILabs ↗
ModelsTogether AI put DeepSeek V4 Flash into production, citing stronger terminal, repository and full-stack coding plus better tool use for long-running agents.X @togethercompute ↗
IndustryPalantir CEO Alex Karp called AI labs' enterprise ambitions "Marxist" after Q2 revenue hit $1.9 billion, up 93%, with $1.1 billion profit.TechCrunch ↗