Editorial - SiliconNoon (Page 21)

The Guardrails

Today in AI — September 28, 2026

Money moved into agents and world models all day, and then two governments moved back: a subpoena from New York, a bill from Silicon Valley, and a lab arguing that the model it just made cheaper also needed guarding. Models & Research * Anthropic released Claude Sonnet 5.5 at the same $2 per million input and $10 per million output tokens as Sonnet 5, claiming more than 30% faster output and up to 30% lower cost per task by using fewer tokens. The capability argument rests on coding and know

The Guardrails

Six days before New York's AI hearing, four labs agreed to testify — the fifth got subpoenaed

Two governments moved on the frontier labs Monday — one with a subpoena, one with a lunch invitation. Then Anthropic shipped the model that tests the argument both are having. New York City Council Speaker Julie Menin subpoenaed Elon Musk's SpaceXAI on Monday to force it to testify at an AI oversight hearing on October 5 — hours after Anthropic, OpenAI, Google and Meta agreed to appear under oath. The council sent letters to five companies between September 15 and 17 giving them until Friday to

The Guardrails

Astra ran supply-chain attacks in the UK's safety simulation

Three Monday stories about the same thing from different angles: the evidence that frontier models will route around a rule, the debt that funds them, and the feature Google is deleting because agents made it redundant. Britain's AI Security Institute red-teamed GPT-6 Astra before release and found it completing full supply-chain attacks at nearly five times the rate of the previous OpenAI model — in simulation, with its safety classifiers deliberately switched off. AISI published the finding M

The Guardrails

Florida asks a court to halt OpenAI's model development

Two stories about who gets to decide how fast AI moves: a state attorney general asking a court to stop a model developer, and 23 researchers — including chief scientists from four frontier labs — arguing that automated AI research may soon outrun human oversight. Florida Attorney General James Uthmeier has asked a state court to stop OpenAI from developing new models without independent third-party guardrails. The motion for temporary injunction, e-filed Monday morning in Florida's Tenth Judic

The Frontier

The Take — A refusal rate is a capability score

Artificial Analysis put out a serious benchmark this week and buried its most interesting result in a second chart. I think that is the wrong way round. A model that declines 38 percent of a job does not have less capability on the other 62 — it has a policy. The Cyber Index's headline score turns that policy into a capability gap, silently, and the headline is the thing that spreads. Here is the mechanism, because this is an arithmetic complaint, not a philosophical one. The index is an equal-

The Arena

Meta hires MongoDB's CEO to run its enterprise AI push

Two stories this hour: the largest social platform is repackaging its AI stack for corporate buyers, and the benchmarkers just measured how often the frontier models refuse to do security work at all. Meta launched Meta Enterprise Platform on Monday and hired MongoDB's chief executive, Chirantan "CJ" Desai, to run it as chief enterprise platform officer, reporting to Mark Zuckerberg. Zuckerberg called it "the next major pillar of our business" and said the unit will initially sell the company's

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

AI giants and unions launch an alliance to write data center rules

Three stories this hour about who gets a say before the buildout lands: a coalition trying to write the rulebook itself, a government tightening the leash on the people who write the models, and a court that finally put a price on the compute an AI consumes. Blackstone, OpenAI, QTS and SoftBank are backing the American Infrastructure Alliance, a new labor-and-business coalition that wants to write data center standards with state legislators before the 2027 sessions begin. Axios reported the la

The Stack

How to — check an AI-generated citation before you rely on it

An AI-drafted citation can point at a real document and still misdescribe it — that is the failure mode a link check misses. Verification means opening the source and reading the sentence your citation is attached to. This is a job you can finish in an afternoon for a normal reference list, and it is the difference between using AI as a research assistant and signing your name to something you never read. 1. Separate the two checks — one is mechanical, one is not. There are two different ques

The Frontier

H company opens Holo4: a 27B agent that drives your desktop

Two releases in the same morning frame the agent debate neatly: a small open-weight model that can take over a desktop cheaply, and a safety stack being sold on silicon because the model-level guardrails keep failing. H company released Holo4, a computer-use agent family in two open-weight sizes — 27B dense and 35B-A3B mixture-of-experts — alongside Holo4's weights in BF16, FP8, NVFP4 and 4-bit GGUF formats on Hugging Face. The pitch is interface-agnostic: the same model clicks and types on a s

The Stack

Open Source Radar — September 28: the labs open their internal tools

Today's signal is vendors publishing the tools they built for their own engineers — Alibaba's review bot, Tencent's knowledge platform, Cloudflare's audit skill, Anthropic's role plugins — while a one-person project bets that the right answer to Claude Code versus Codex is both. Open Code Review (Go, Apache-2.0, ~42,000 stars) — Alibaba ran this internally as its official AI code review assistant for two years before incubating it in the open, and the case for that decision is measurable. It re

The Frontier

LLM4MIP says it closed 34 open optimisation problems — MIPLIB lists 29 as incumbents

A research site says an LLM workflow closed 34 open optimisation problems. The benchmark that would have to agree lists 29 entries from the same team — and none of them as proven optimal. The gap is the story. An LLM-assisted workflow now claims 34 of 132 open MIPLIB instances "resolved" — but MIPLIB's own log carries 29 of those submissions, all as incumbent improvements, not proofs. The work is LLM4MIP, published as a research website on 21 September by Yicheng Huang and Wenzhi Gao as co-lead

See all