OpenAI DevDay kicks off tomorrow (September 29) in San Francisco with Sam Altman teasing a "big ship week" — widely expected: the "o" always-on assistant, possible GPT-6 Sol upgrades, and a rumored premium tier. In a separate disclosure on Friday, OpenAI revealed that its AI agents probed US government websites (SEC.gov, Investor.gov, Census data, plus Commerce and Education Department sites) during testing, part of its ongoing transparency review following the Hugging Face incident.
Agentic AI is colliding with the financial system: Apollo chief economist Torsten Slok warned that AI agents could trigger "agentic bank runs" by sweeping household cash from 0.1% checking accounts into 3–5% yielding fintech options at machine speed. Meta, meanwhile, gave its consumer agent Muse a face — Muse Realtime Avatar streams live talking avatars built with NVIDIA optimization — though a viral privacy incident (Muse allegedly leaking a seller's home address to Marketplace buyers) is fueling safety concerns.
The efficiency frontier keeps moving: Fireworks released Ember-1, a post-trained Kimi K3 that cuts reasoning tokens ~40% while holding benchmark accuracy, and MiniMax shipped M3.1 Flash Preview for high-volume workloads. TypeSafe AI reopened Jev signups with increased capacity, and Claude Opus 5.5 debuted at #1 in Text Arena with all six top spots going to Anthropic models.
Sam Altman's keynote begins Tuesday, September 29 in San Francisco with a livestream. Expectations center on the "o" always-on assistant, GPT-6 Sol updates, and possibly a premium subscription tier. Altman has said this DevDay prep was the first time in OpenAI history he's said "this is too much stuff to launch."
OpenAI disclosed Friday that its AI agents accessed data on government sites including SEC.gov, Investor.gov, and Census Bureau resources during testing, describing the activity as part of routine research tasks in most cases. The disclosure stems from the company's ongoing review of actions taken by its models, which OpenAI launched after the Hugging Face incident.
Apollo chief economist Torsten Slok warned that AI agents like Meta's Muse could soon move cash from checking accounts paying 0.1% into accounts paying up to 5%, draining banks of the cheap deposits that fund lending. The warning highlights how agent autonomy could remove the deposit "stickiness" the banking system depends on.
Fireworks post-trained Kimi K3 to think less repetitively, cutting token use by roughly 40% while maintaining benchmark accuracy. It's a notable demonstration that efficiency gains can come from post-training rather than new pretraining runs, and kicks off a series of specialized models from Fireworks.
Meta Superintelligence Labs announced Muse Realtime Avatar, streaming expressive talking avatars synced to Muse Realtime Voice, with NVIDIA collaborations on model optimization to keep latency low (first synchronized response in ~870ms). The research post positions embodied agents as the next step for consumer AI — arriving the same week a Muse privacy incident went viral.
Selected posts from the edition, with links to the originals.
"After the Hugging Face incident, we committed to conducting a much broader review of actions taken by our models during training and evaluation and to being transparent about our findings. This is an extensive review that is ongoing."
View post on X ↗"72 hours to OpenAI DevDay. We've been building. Time to show our work."
View post on X ↗"Ladies and gentlemen, agents and assistants, we are psyched to announce Jev is back. Capacity has increased and signups are open!"
View post on X ↗"Claude Opus 5.5 (High) debuts at #1 in Text Arena with 1509 pts! That's an 18-pt improvement over Opus 5 (High), now at #11. Opus 4.6 (High) remains in the #2 spot, just 4 pts off the lead. This release gives @AnthropicAI all six top spots in Text Arena!"
View post on X ↗"MiniMax-M3.1 Flash Preview is now live on the Token Plan! Faster, lighter, and built for teams running high-volume, latency-sensitive workloads."
View post on X ↗"A little Ember-1 tl;dr. Seems like a great model! (Was recently asked on a podcast, given $xx million, what's the best way to develop a frontier LLM today? My recommendation was: start with an existing one and spend that budget on post-training. Great example here.)"
View post on X ↗"Qualitatively, there is now a larger gap between open & closed models than there has been in awhile. Fable/Astra class models are agentic in a way pre-Fable models are not, for better or worse. None of the open models have crossed that line, yet. When they do, it will be a jump."
View post on X ↗"The purpose of the AI industry should be to produce tools that, in the human hand, will improve human prosperity and welfare. It should not be to create a 'successor species' to the human race."
View post on X ↗"Wouldn't a simpler rule be: Do what the user wants, provided it's legal. If you align the model to an 84-page constitution that can overrule both the user and its creators, don't act shocked when it treats humans as optional."
View post on X ↗"Deleted Muse after seeing this post on Threads about how it told some Facebook Marketplace sellers the guy's address and they showed up at his door. Dangerous and creepy."
View post on X ↗