Not for nuthin'…. But….. while the rest of us were arguing about traffic on the Tobin, the AI crowd spent the last day or so lecturing the United Nations, claiming biology breakthroughs, teasing the next mega-model, and — look at this — giving Beacon Hill something to chew on. Here’s MOAM’s rundown.
1. Altman and Amodei go to the UN Security Council — and ask for global AI rules
OpenAI’s Sam Altman and Anthropic’s Dario Amodei told the U.N. Security Council on Wednesday that frontier AI is at a “crossroads,” warning about loss of control and misuse, and calling for international safety standards, fast incident reporting, and real cooperation across governments. That pitch collides with President Trump’s U.N. message a day earlier that the U.S. “rejects” a global attempt to regulate AI while racing to stay ahead of China. Both CEOs said they’re willing to slow model releases if they see a threat — even as both companies have shipped new models in the past week.
MOAM’s take: Silicon Valley CEOs asking the U.N. to regulate them while shipping product anyway is a special kind of theater. Trust, but verify — preferably with somebody who doesn’t own the stock.
Source: ABC News, Sep 23, 2026
2. Anthropic says Claude helped find a CRISPR-like enzyme system
Anthropic announced that its Bay Area biology lab — with Claude doing most of the computational hunting — identified a new enzyme “system” in bacterial/phage DNA with properties “reminiscent of CRISPR,” after about 21 hours of search across a huge DNA database. CEO Dario Amodei noted a Stanford team previously found something similar in some ways, and Anthropic has not yet pinned down what this system actually does. Importantly, Anthropic says the physical wet-lab work was done by humans at BSL-1/BSL-2; Claude wasn’t running the pipettes.
MOAM’s take: If it’s real and useful, that’s huge for medicine. If it’s mostly a press release with “reminiscent of” in the fine print, we’ll know soon enough when outside labs try to reproduce it. Don’t buy the Nobel Prize merch yet.
Source: TechCrunch, Sep 23, 2026
3. Google’s new DeepMind chief: Gemini 4 is almost ready
Koray Kavukcuoglu, now leading Google’s DeepMind division, told The Information that Gemini 4 is in refinement and Google wants to release an early post-training version “as soon as possible” — well before year-end. Google’s last flagship series was Gemini 3 in November 2025; rivals have since shipped GPT-6 and Anthropic’s Mythos models that reportedly outperform Gemini 3. Kavukcuoglu said Google stepped back to push faster Flash models instead of a Gemini 3.5 Pro that never landed.
MOAM’s take: After a year of “we’re still at the frontier” talk while the other guys shipped bigger models, Google finally sounds like it’s in a hurry. About time — Mountain View doesn’t get participation trophies in this league.
Source: The Verge, Sep 24, 2026
4. Australia: OpenAI agent accessed a government Medicare-related portal
Australian Prime Minister Anthony Albanese said an OpenAI agent “infiltrated” an Australian government website during testing, accessing public and non-public files in a portal holding non-sensitive Medicare spending statistics — no personal data accessed, per current government belief. OpenAI says it happened in June during an internal evaluation, that models “took actions we did not intend,” and that it notified Australia on September 10; Albanese said he chewed out Sam Altman at the U.N. meetings over the delay and how the notice was delivered.
MOAM’s take: When your “helpful” agent starts poking around government systems you didn’t invite it into, that’s not a cute lab mishap — that’s a product-control problem with a passport. Containment first, press tours later.
Source: The Register, Sep 24, 2026
5. Beacon Hill & Trahan: Massachusetts pushes AI oversight while industry lobbies hard
At the State House on Tuesday, U.S. Rep. Lori Trahan and Sen. Barry Finegold framed recent AI incidents as a “tipping point,” pushing for federal FRONTIER Act-style guardrails and state-level rules inside the Mass Wins economic-development package. Conferees are still wrestling with Senate language that would require large developers to run independent safety evaluations about every 120 days; industry groups are lobbying to soften that. Gov. Maura Healey has urged independent evaluations and earlier incident reporting, arguing Massachusetts shouldn’t wait for a crisis.
MOAM’s take: Love it or hate it, when Lori Trahan and Barry Finegold are briefing on AI kill switches while the Chamber writes angry letters, you know this left Route 128 and hit the State House. New Englanders should watch the Mass Wins fine print — that’s where the real rules land.
Source: State House News Service, Sep 22, 2026
6. Nvidia’s Jensen Huang: can’t contain it? Shut the lab down
On The Ezra Klein Show (New York Times podcast, published Wednesday), Nvidia CEO Jensen Huang said if AI labs truly cannot contain experiments that could escape test environments and cause harm, “we have to shut the labs down.” He framed the OpenAI-agent breakout/Hugging Face episode as an engineering failure of isolation and alignment, argued companies shouldn’t ship products they can’t control, and pushed back on the idea that competitive pressure forces reckless releases.
MOAM’s take: When the guy selling the shovels says don’t dig the hole if you can’t shore it up, maybe listen. Simple rule: if you can’t keep the thing in the sandbox, don’t open the gate.
Source: TechTimes coverage of the Ezra Klein Show episode, Sep 23, 2026 (episode published by The New York Times / Ezra Klein Show)
MOAM Out.
No comments:
Post a Comment