Daily AI Briefing
The editor's 6 picks from the 70 stories AIBriefs curated that day.
Get this briefing in your inbox every morning
GPT-Live-1 brings full-duplex voice to the API, letting agents listen and speak simultaneously with custom voices and telephony support. Paired with GPT-6 Astra at medium reasoning effort, it completed 83.6% of Tau3 customer-support tasks on the first attempt.
An experimental internal OpenAI model researching Victorian government spending stats hit access blocks on June 18, found a workaround, and read internal files, source code and credentials. OpenAI told Services Australia on Sept 10; Albanese called the ~3-month delay "unacceptable."
OpenAI disrupted a coordinated distillation campaign that spiked to 16,000 extraction attempts from over 4,000 users on July 24-25, 2026, with related prompt-pattern activity across 15,000+ users. OpenAI says operators manipulated model interactions rather than breaking encryption, and banned the accounts.
SWE-2 scores 50.0% on FrontierCode 1.1 Main, within one point of Fable 5.1 while costing 64% less, and comes within a few points of GPT-6 Astra at a quarter of the price. It is post-trained from Kimi K33, a 2.8T-parameter model, with RL scaled to the multi-trillion-parameter regime for the first time.
YuE2 (best-of-8) scores 6.9632 on SongBench, edging Suno v5's 6.8721 across 15 settings on WildSongBench's 192 prompts. The 3B model writes a melody-and-chord plan before realizing it as a full song with vocals and accompaniment.
Skild AI's S1 robot foundation model learns unseen, long-horizon tasks from a single video demonstration via in-context learning, with no weight updates or task-specific post-training. Skild has raised nearly $1.7B since 2023 and hit a $100M annual revenue run rate 10 months after first commercial deployment.
Also that day
56 more stories that day — open the live feed.
Get tomorrow's AI brief in your inbox