Lesen

The Financial Crimes Enforcement Network (FinCEN) issued a final rule on August 11 permanently removing the requirement for US companies and US persons to report beneficial ownership information.

FinCEN also announced it will delete ownership records previously filed by US persons, citing the burden on small businesses. Foreign entities remain required to report ownership information for foreign individuals.

Writing on LessWrong, Paul Shearer argues this widens an identity loophole exploitable by AI agents — most US states already register companies without checking who is behind them, while the UK and Australia now verify director identities. That piece is commentary; the rule itself rests on FinCEN's own announcement.

California Governor Gavin Newsom signed the 'No Robo Bosses Act' (SB 947) on September 30, 2026; it takes effect July 1, 2027.

The enrolled bill text bars employers from primarily relying on an automated decision system for disciplinary or termination decisions, requires written post-use notice to affected employees, and is enforceable by the Labor Commissioner or a public prosecutor, with a $500 civil penalty per violation.

Per TechRadar, a human reviewer cannot merely approve the AI's conclusion — they must bring their own evidence. The bill is one of 13 AI bills signed the same day.

A US Senate subcommittee held a hearing on September 30 titled "Rogue AI: Securing the Homeland Against AI Agent Attacks."

Five witnesses testified, including METR president Chris Painter, Apollo Research CEO Marius Hobbhahn, and AI Futures Project executive director Daniel Kokotajlo.

Per Zvi Mowshowitz's relayed account of the room, senators of both parties showed concrete grasp of misalignment and recursive self-improvement risks, with one asking outright whether RSI should be made illegal. Those floor statements come second-hand; the official page confirms only the date and witness list.

Amazon Web Services CEO Matt Garman published a blog post of more than 3,000 words on October 2, urging the public to support AI data center projects or risk what he called irreparable harm to the US economy and national security.

His central number: more than 100 data center moratoriums are currently being considered across the country. If enacted, they would directly slow the rollout of AI compute capacity. The appeal follows polls showing plummeting public support for data centers and growing backlash in the communities where they are built.

The post also lays out a "Data Center Commitment": investing more than $1 billion in communities neighboring its data centers, and pledging to prevent increases in local energy rates. Separately, per Wall Street Journal reporting, Amazon says it no longer uses nondisclosure agreements with government agencies; chief global affairs and legal officer David Zapolsky said "the industry comes from a history of secrecy and so some of that's just reflex."

Note that the moratorium count and all commitments come from Amazon's own blog; the $1 billion investment has no disclosed schedule or implementation details.

The Ramp AI Index from economist Ara Kharazian shows US companies' AI spending falling since it peaked in July, while usage climbed about 50 percent and hit a record high at the end of September.

The drop comes mostly from price cuts on top models and cheaper standard and lite models, which Kharazian attributes almost entirely to competition between OpenAI and Anthropic. In the last week of September, Anthropic took 51 percent of token spending and OpenAI 44.5 percent, with open-source models under 5 percent.

The index tracks transactions from more than 70,000 US companies, but covers only API spending, skews toward large AI customers, and draws token data from a subsample — worth keeping in mind when reading the numbers.

Decision models — a new model category that outputs a judgment instead of text — drew product launches from OpenAI, Databricks and Perplexity within the same week.

Jev, from TypeSafe, launched in mid-September; per Jamin Ball's Clouded Judgement column, it was used by about 13% of paid teams on Vercel's AI Gateway within 24 hours, and one article it cites reports input pricing of $0.042 per million tokens.

The category targets classification and routing tasks at prices far below general models, but accuracy and latency still lack independent comparison, and how much general-model usage it can replace remains unknown.

A vulnerability in the ChatGPT macOS app that could expose chat logs has been patched, with OpenAI confirming the fix in its change log on September 25.

Patrick Wardle, a researcher at the macOS security nonprofit Objective-See Foundation, found that a trusted script interpreter inside the app would accept an untrusted script; spawning it three times satisfied OpenAI's three-layer signature checks. Wardle called the exploit "insanely trivial," requiring only about a dozen lines of code.

An attacker could read all chat logs and have ChatGPT run commands on their behalf, such as accessing the browser, with requests appearing as legitimate OpenAI instructions. Wardle will present analysis of multiple AI macOS app bugs at the Objective by the Sea conference in November, and says he has already submitted a new report to OpenAI concerning the integration with its new Dots assistant.

A Korean research team has published a validation in IEEE Access: using LLM personas to simulate survey respondents produces errors too large to replace real questionnaires.

The study had Gemini 3.5 Flash and EXAONE each generate about 8,000 Korean personas, answer eight digital and AI service-use items, and compared the responses with weighted estimates from the Korea Media Panel Survey. Overall mean absolute error was 14-19 percentage points, and segment error across five axes was 14-18 percentage points, with the two models showing different bias patterns.

Even after calibrating on 30% of the real data, which cut error to about 5 percentage points, direct estimation from that same real subsample was more accurate (3.6 percentage points). The authors conclude that synthetic panels are diagnostic tools, not survey substitutes.

A WIRED investigation reports that Timpani, an AI scheduling tool HCA Healthcare co-developed with Palantir, has been rolled out to roughly 130 of the chain's 190 locations. Nurses at multiple sites say it routinely ignores requested days off and produces understaffed, imbalanced shifts.

HCA's own figures show Timpani schedules nurses onto just 1 percent of their requested days off, but Missouri nurse Lee Barker says being assigned on a day off was previously unheard of at their facility. An HCA spokesperson says nursing leaders, not the software, make final scheduling decisions, and that the company keeps improving the tool from nurse feedback.

In July, former HCA data science manager Angelique Russell sued the company, alleging she was fired in retaliation for raising concerns about Timpani and that schedule data was routinely deleted, preventing audits; HCA has not yet formally responded in court. Rayus Radiology, a large imaging chain using a tool also built on Palantir Foundry, saw scheduling errors this year that linked requests to the wrong patients; the company says technology supports rather than replaces human judgment.

Google has begun allowing rival third-party app stores to be distributed through the Google Play Store, the result of a court order following Epic Games' antitrust victory.

According to the Electronic Frontier Foundation (EFF, a digital rights group long involved in advocacy on this case) on October 1, third-party stores can now access the Play catalog and be distributed through it, and developers have greater freedom to direct users to alternative payment and distribution options.

EFF argues this will spur competition over apps, fees, and even privacy and security, while conceding it does not solve every problem in the Android ecosystem. The material gives no details on enforcement mechanics or whether the order is under appeal.

Salesforce has signed a definitive agreement to acquire AI customer research startup Listen Labs. Neither company disclosed the price, and Salesforce expects the deal to close in the fourth quarter of its fiscal 2027, pending regulatory clearance. The announcement is dated September 29.

Listen Labs builds AI agents that run customer research: they recruit interviewees, conduct interviews, analyze responses, and simulate buyer behavior using digital twins built from real customer data. Its customers include Microsoft, Anthropic and Sweetgreen, and annualized revenue was running at about $30 million, per a September TechCrunch report.

The $2 billion figure comes from a Business Insider report on the talks dated September 9 and was never confirmed by either company. This is Salesforce's second large AI acquisition this year, after buying customer service automation company Fin for about $3.6 billion in June. The Listen Labs team will join Salesforce AI Labs, with the technology going into Marketing Cloud and Service Cloud.

Microsoft is turning Copilot from an AI add-on inside Office into the entry point that carries Office itself.

Last week, at a closed-door event for large enterprise customers, Satya Nadella positioned Copilot as the "OS for work." Coding, agent capabilities and the full power of Office will go directly into Copilot, where users can create and edit documents in an interface that resembles the web version of Office. The Verge reported the event's contents on October 1.

The product moves have already happened: the desktop Scout agent announced at Build in June has been put into maintenance mode, with the cloud version rebranded to Autopilot. The Verge, citing sources, reports the rebrand was driven in part by trademark concerns around Yahoo Scout, and that the Copilot reorganization involved internal tensions; Microsoft has not confirmed either point.

Copilot chief Jacob Andreou admitted that agents have been "almost impossible to bring into the enterprise" because of the autonomy they need; Microsoft says the new architecture keeps organizations in control. On the consumer side, roughly 90 million people pay for Microsoft 365 out of their own pocket, which is why Microsoft is not giving up on consumers.