AI Frontier · Aug 15, 2026
The 7-day window yielded 20 items in the
ai_frontiercategory and 0 inai_security(the three sources HiddenLayer / Lakera / Protect AI have no RSS and are not yet implemented in the fetch layer — a known gap, not evidence that there was nothing to report). 11 items selected, ordered most recent first, with items from the last 48 hours at the top; the remainder were enterprise case studies and marketing content, skipped as low-signal.
Last 48 Hours (Aug 13)
Google DeepMind releases Gemini 3.7 Flash (Aug 13, 17:04) A lightweight model iteration aimed at low-latency use cases. DeepMind Blog
Google Sheets adds "Sheets canvas" (Aug 13, 16:45) Turns spreadsheet data into interactive dashboards, study trackers, and more via prompts. Google Blog
OpenAI publishes the GPT-5.6 builder's guide (Aug 13, 11:00) Aimed at startups, covering how to cut costs and build faster through smarter model selection and new Responses API capabilities. OpenAI News
OpenAI previews Ultrafast mode: GPT-5.6 Sol up to 14x faster (Aug 13, 10:00) A new API service tier built on Cerebras compute, with output speeds reaching up to 750 tokens per second. OpenAI News
OpenAI names Dali Rajic Chief Revenue Officer (Aug 13, 09:00) Leading the global revenue organization. OpenAI News
Aug 10 — Aug 12
OpenAI expands Daybreak: the cyber defense window is narrowing (Aug 10) Introduces GPT-5.6-Cyber, available to authorized Daybreak Red users for authorized vulnerability research, exploit validation, and security testing. This is the item in this edition most relevant to this project's focus. OpenAI News
OpenAI: putting frontier cyber models in more trusted hands (Aug 10) Vetted Daybreak partners can now use OpenAI's frontier cybersecurity models to deliver governed, authorized security services to their customers. OpenAI News
Daybreak models arrive on AWS (Aug 11) OpenAI partners with AWS to deliver Daybreak cybersecurity capabilities through Amazon Bedrock in support of enterprise security workflows. OpenAI News
Google DeepMind: putting sign language AI into users' hands (Aug 12) Introduces a sign-language-to-text (SL2T) model serving deaf and hard-of-hearing users. DeepMind Blog
Google's AMIE medical AI system demonstrates real-time clinical video consultation (Aug 11) The first study of real-time video consultation, conducted in simulated scenarios. Google AI Blog
OpenAI research: how enterprises are moving AI from assistance to execution (Aug 12) Details the current state of enterprise agentic AI adoption, with frontier firms continuing to lead on ChatGPT/Codex usage. OpenAI News
OpenAI begins testing ads in ChatGPT (Aug 11) Intended to sustain free access; OpenAI emphasizes clear labeling, answer independence, privacy protection, and user control. OpenAI News
Known Gap
ai_security (HiddenLayer Research / Lakera Blog / Protect AI Blog) has been at 0 items continuously — these three sources have no RSS/Atom feed and no autodiscovery link on their pages, while the fetch layer currently handles only rss/atom and adapted JSON sources, skipping anything with feed_type=html. This is a known limitation of the fetch architecture; it does not mean there was nothing relevant this period, and it should not be treated as a bug to fix.