Morning Edition · Tuesday, August 11, 2026
Tech
OpenAI Ships a Purpose-Trained Offensive Security Model to Vetted Partners Only
GPT-5.6-Cyber answered 95 percent of advanced security prompts in OpenAI's own testing, and reaches customers solely through an identity-verified Daybreak Red tier that will require a hardware key from September 1.
Tech
Meta Superintelligence Labs Returns to Open Weights With a 30B Agent Model Under Apache 2.0
Muse Glimmer is dense rather than mixture-of-experts, carries a context window above 128,000 tokens, and fits under 20 gigabytes at four-bit quantization.
Tech
Hassabis Leaves the DeepMind Chief Executive Role as Jeff Dean Departs Google
Koray Kavukcuoglu takes over daily operations of the lab behind Gemini, while Hassabis becomes Alphabet chief scientist and concentrates on Isomorphic Labs.
Tech
Agent Tooling Fights Its Own Token Overhead as Qwen Ships a Harness-Agnostic Multimodal Layer
One open-source client claims a 97 percent cut in the tokens spent discovering tools, a cost that scales with every server an agent connects to.
Tech
Three New Papers Attack the Weak Points of Multimodal Models at Once
Researchers target the security of optical token compression in DeepSeek-OCR, replace static hallucination benchmarks with fuzzing, and catalogue how modality fusion opens new attack surface.
Tech
Researchers Test Whether Claude Models Write Code That Passes a SOC 2 Audit
The study checks encryption, credential handling and regulated-data storage across four use cases and three model versions, including Opus 5.
Tech
Schmidt Puts Frontier Training Runs at $200m to $500m With a Six-Month Useful Life
The former Google chief executive says he sees no evidence scaling has stopped, a position that raises rather than lowers the capital intensity of staying at the frontier.
Markets
Texas Halts New Data Center Grid Approvals as Requests Reach 474 Gigawatts
OpenAI wrote to Governor Greg Abbott pledging to pay for its own electric infrastructure and limit water use, days after Abbott ordered a statewide grid and water audit.
Tech
Sanders Tells Three AI Chief Executives to Pause Development or Face the Senate
The letter cites the Stanford and Arc Institute study published in Science on August 6 in which generative models designed 16 working viruses that do not exist in nature.
Tech
A 33-Billion-Parameter Video and Audio Model Now Runs on a MacBook, Slowly
Salvatore Sanfilippo has written a native Metal inference engine for MiniMax-H3, while a separate MLX port needs about 115 gigabytes of weights and takes just under 45 minutes per clip on an M5 Max.
Tech
Google Puts Agents Inside Ads and Analytics as OpenAI Documents Its Own Finance Automation
The distribution channel for enterprise agents is shifting from developer APIs to the products finance and marketing teams already use every day.
Tech
Meta's Open Vision Models Cut a Month of Beamline Analysis to Roughly 15 Minutes
SAM 3 and DINOv3 run on 300 Nvidia A100 accelerators at a Department of Energy computing center, segmenting X-ray and neutron imaging in real time for a flagship Genesis Mission project.