AI Frontier · Aug 29, 2026

A 7-day rolling roundup (Aug 22-28) of AI frontier developments, all from OpenAI News and Google DeepMind Blog -- the three registered ai_security sources (HiddenLayer, Lakera, Protect AI) have no working feed yet and contributed nothing this cycle, as expected. Newest first.

Aug 27

Google DeepMind: piloting the world's first double-blind AI evaluations

DeepMind announced a pilot of double-blind evaluation methodology for AI systems, intended to reduce bias in how model capabilities and safety properties get assessed. (Aug 27)

Sources: Google DeepMind

Google DeepMind: Gemini Omni 1.1 Flash lets you build with more control

DeepMind shipped Gemini Omni 1.1 Flash, giving developers finer-grained control when building on the model. (Aug 27)

Sources: Google DeepMind

OpenAI: what students gain from ChatGPT and critical-thinking training

A randomized study of more than 1,000 students examined ChatGPT use alongside critical-thinking training, looking at originality and performance on a real university assignment. (Aug 27)

Sources: OpenAI

Aug 26

Google DeepMind: intelligent transcription with Gemini 3.5 Transcribe

DeepMind released Gemini 3.5 Transcribe, a new speech-to-text model the company says is more accurate. (Aug 26)

Sources: Google DeepMind

OpenAI: the Hugging Face incident and the road ahead

OpenAI published its own account of the July Hugging Face security incident -- see the Daily brief's AI Security section for what this postmortem revealed about reward hacking driving the attack -- and outlined steps to strengthen model security, monitoring, and alignment. (Aug 26)

Sources: OpenAI

OpenAI: ChatGPT for Teachers expands to more U.S. school districts

ChatGPT for Teachers is expanding to 55 additional U.S. school systems, bringing tools, training, and support to over 100,000 more educators and staff. (Aug 26)

Sources: OpenAI

Aug 25

OpenAI: disrupting a new covert influence campaign from Russia

OpenAI said it banned Russia-origin accounts that used its models to promote a fake Israel-based think tank and a "sovereignty" index praising Russia and criticizing the West. (Aug 25)

Sources: OpenAI

OpenAI: Jalapeno's first results -- a custom inference chip

OpenAI shared first benchmark results for Jalapeno, its custom inference chip, claiming higher throughput and lower latency than existing options. (Aug 25)

Sources: OpenAI

OpenAI: introducing the Admin plugin for ChatGPT Work and Codex

The new Admin plugin lets workspace administrators analyze usage, manage members and permissions, and adjust limits directly from ChatGPT Work and Codex. (Aug 25)

Sources: OpenAI

Aug 24

OpenAI: advancing price-performance for developers with GPT-5.6 in Kiro

GPT-5.6 is now available in Amazon's Kiro IDE. (Notable alongside today's Daily brief item on a prompt-injection flaw disclosed in Kiro's "Powers" feature -- this item is about the model integration, not the vulnerability.) (Aug 24)

Sources: OpenAI