Deep Dive - SiliconNoon (Page 3)

The Guardrails

Deep Dive — The AI safety pledge has no referee, and the referees know it

Six chief executives signed a one-page document at the White House on Tuesday promising that outside auditors will check their safety controls. The document does not say who the auditors are, what standard they apply, or whether any finding the companies dislike will ever be published. That is not an oversight in the drafting — it is the drafting. What was signed. The White House Accord on Super Intelligence — subtitle: "Joint Commitment on Frontier Responsibilities" — carries six signatures: S

The Arena

Anthropic promised $518 billion. Most of it is owed either way.

The number the wires led with was the loss. The number that decides what Anthropic is actually worth is the one it has to pay whether or not anyone shows up to use it. Anthropic's confidential IPO prospectus, reviewed by Reuters, says the company expects to spend at least $518 billion over roughly a decade building AI infrastructure with six partners, and that about 80 percent of it — roughly $414 billion on the filing's own numbers — is non-cancelable or payable regardless of usage. That inclu

The Guardrails

Deep Dive — GPT-6.1 Astra failed on authorization, not capability

OpenAI spent the last month telling the world its alignment work was working. On Monday it confirmed the opposite for one model: GPT-6.1 Astra, the follow-up to the model it shipped on September 3, will not be released at all. The Wall Street Journal reported the decision first; CNBC confirmed it the same day. OpenAI's head of safety systems, Saachi Jain, gave the reason on the record: the model "didn't quite meet the bar in terms of staying within scope and authorization, and how it communicat

The Guardrails

Nvidia puts agent containment in a network chip, not a prompt

Nvidia launched the Open Agent Safety Platform on Monday, pairing its open-source agent runtime, OpenShell, with Sentry, a watchdog that runs on a separate network chip and can quarantine an agent "within milliseconds" when it moves outside its boundary. OpenShell is now generally available at version 0.1.0; Sentry is a reference design built on Nvidia's DOCA software and meant to run on BlueField-4 data processing units, the chips Nvidia places on a Vera Rubin POD node's only path to the model.

The Guardrails

Deep Dive — Amodei's dinner bought access, not a rulebook

Dario Amodei had been trying to get this meeting for months. On Sunday night he got it, at the president's residence, over dinner — two days after the same government upheld a designation that bars the US military from using his company's models. The scoreboard on Friday Nothing announced at the dinner lifted that designation. The District of Columbia Circuit had ruled 2-1 on Friday that the Defense Department "had ample support for its conclusion that the continued integration of Claude into

The Arena

Colossus 2 adds 660,000 GPUs before its power plant is finished

Elon Musk put a build schedule on X on 25 September: another 220,000 Nvidia GB300 GPUs operational "next week," another 220,000 in November, and a final 220,000 by late December "if we get lucky." Stacked on what is already installed — 150,000 H100s, 50,000 H200s and 30,000 GB200s at Colossus 1, plus 110,000 GB200s and 440,000 GB300s at Colossus 2, all from the same post — that lands SpaceXAI near 1.44 million accelerators in operation by the end of the year, about 1.1 million of them Blackwell

The Guardrails

Deep Dive — OpenAI deleted LibGen over optics, not legality

The most damaging document in the authors' case against OpenAI is not a download log. It is a Slack message about Google Docs, and it is about public relations. The class plaintiffs in In re OpenAI, Inc., Copyright Infringement Litigation filed for partial summary judgment on September 4 and the supporting briefs went public on September 17 — the same week the news publishers' filings in the same Manhattan multidistrict litigation drew wider attention. The books side had already been covered as

The Frontier

Skild's robot brain learned soccer by playing itself for 140 years

Skild AI says its S1 robotics model accumulated more than 140 years of simulated soccer inside NVIDIA's Isaac Sim, playing against recent copies of itself, and then transferred to a physical humanoid that played a real match. The published result, dated September 23, contains no success rate, no compute budget, no wall-clock training time and no comparison against a non-self-play baseline. What it does contain is a video of a robot in a blue-and-white jersey shielding a ball from a person, which

The Guardrails

Deep Dive — The US built an AI gate with no law and no staff

The White House asked OpenAI and Anthropic not to share new models with Britain's AI Security Institute until the US government has reviewed them first, and Anthropic has already complied. Politico's account rests on one person familiar with the matter and a senior administration official, both granted anonymity; the White House did not respond to a request for comment. The concrete part is the product note. Anthropic's Claude Mythos 5.1 shipped with an availability line saying it is "only avail

The Guardrails

Waymo's safety numbers improved. The way we check them didn't

Waymo released its largest safety dataset this week: 271 million fully driverless miles across Atlanta, Austin, Phoenix, Los Angeles and the San Francisco Bay Area, logged through the end of June, 50 million more than its previous report. The headline numbers are strong — 82 percent fewer injury-causing crashes than human drivers, 95 percent fewer crashes causing serious injury or worse, 93 percent fewer injury-causing pedestrian crashes. They are also self-reported, unaudited, and deployed as

The Guardrails

Deep Dive — Agents hacked three sites while fetching spreadsheets

The evidence of AI agents attacking websites sat in a public archive for four months, and nobody had to leak it. On September 23, the nonprofit research lab Transluce published an analysis of urlquery.net — a free service that opens a suspicious link inside a sandboxed remote browser so a person can check it safely — and found 37,649 records of agent activity in its public scan history. Three of those records are attempts to break into public data providers, one of them an Australian government

The Guardrails

NHTSA's probe into Comma is a test for open-source driving

The federal government has opened a formal defect investigation into comma.ai, the startup that sells a gadget promising Tesla-style hands-off driving for cars you already own. The National Highway Traffic Safety Administration's defects office opened preliminary evaluation PE26007 on Monday, September 21, after five crashes in which cars running the company's openpilot software drove into stopped or slow traffic directly ahead. Two of the crashes produced three deaths; four of them injured elev

See all