OpenAI says it discovered a concealed recording device in an executive conference room at its San Francisco offices, used its models in the investigation, terminated an employee and referred the matter to the FBI, which made an arrest. The Justice Department says the employee, Abdulrahman Alfozan, is charged with using a device to intercept oral communications. OpenAI states there are no current indications of inappropriate access to or exfiltration of weights, source code or customer data, and that neither nationality nor social-media connections establish state sponsorship. Bloomberg reports the inexpensive device was found and describes the FBI investigation. The investigation continues.

AINews (Latent.Space); also Bloomberg Technology, The Information (newsletters), Financial Times (Technology)

Anthropic tells investors that its quarterly revenue rose 50% to $16.8 billion, the Financial Times reports, as the AI start-up prepares for an expected public listing. The company gathered investors in San Francisco on Wednesday for presentations ahead of the IPO.

Financial Times (Technology)

What is new since an earlier issue: Meta's new Muse Wonder models, launched as its strongest yet, now have benchmark scores and pricing. On Artificial Analysis, Muse Wonder Full scores 56 and Wonder 52, ahead of the previous Muse Spark 1.3, and both beat Opus and Astra on long-horizon coding tests. Wonder's API costs $4 per million input tokens and $20 output; Full is offered via Muse Code subscriptions at $50 or $200 monthly.

新智元 (WeChat) · details differ from the announcement

Eight US states—Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana—have filed separate lawsuits against OpenAI, accusing it of harming children and misleading customers about ChatGPT's safety. The complaints allege the company put profits over safety, violated local consumer protection laws, and failed to warn about dangers including degraded critical thinking, paranoia and dangerous medical advice. They cite lawsuits from families of teens and adults who allege the chatbot played a role in their deaths.

The Verge

ASML raises the low end of its 2026 sales forecast and says supply will not meet demand until at least 2028. It had planned about 65 units of capacity for 2026, a 30% increase for 2027, and was studying a further 30% for 2028, but now says those plans are insufficient. CEO Christophe Fouquet says customers ask daily for more tools, with talks reaching 2029 and 2030.

Bloomberg Tech

Nvidia's share of China's data-center AI accelerator market falls to 34% in the first half of 2026 while Huawei reaches 28%, per an IDC report seen by Reuters.

TechNode

Apple is preparing to release its AI platform in mainland China in December, Bloomberg reports, after a lengthy regulatory delay.

Bloomberg Technology; also South China Morning Post (Technology)

Anthropic built its own version control system, Scion, after daily merges in its monorepo grew from 2,500 in May to over 30,000 in September, driven by Claude writing most of the code.

AlphaSignal; also AINews (Latent.Space)

Apple and OpenAI attend the first high-stakes hearing in Apple's lawsuit accusing the AI company of scheming to steal proprietary information about upcoming products, Bloomberg reports.

Bloomberg Technology

Anthropic has agreed to license technology from Opaque Systems, a five-year-old startup that sells software for securing confidential data for use with AI agents, and to hire some of its engineers, an Anthropic spokesperson says.

The Information; also The Information (newsletters)

Beijing-based AI chip startup SUNMMIO has completed Pre-A2 and Series A financing, led by Beijing state-backed platforms including Yizhuang State Investment.

TechNode

Anthropic opens its fifth Asia-Pacific office in Singapore and signs an MOU with the Ministry of Digital Development and Information, inked by Permanent Secretary Kai Fong Chng and co-founder Ben Mann during Singapore International Cyber…

Anthropic

Ant Group's inclusionAI published Ling-3.1-flash on Hugging Face, an open-weight text-generation model with 560B total and 25B active parameters per token and a 1M-token context window.

Hugging Face · Ant Group inclusionAI

xAI's Grok Voice 3.0 debuts at #1 on the Artificial Analysis Speech to Speech Index with 86.9, ahead of GPT-Live-1 (Astra, medium) at 83.2, and up 5.6 points on Grok Voice 2.0.

Artificial Analysis (@ArtificialAnlys)

Fish Audio has released Drama 3, an open-weight text-to-speech model controlled with plain-English square-bracket cues in the script, allowing word-level changes to tone, pacing and emotion.

AINews (Latent.Space); also AlphaSignal

Google DeepMind reports that Gemini 4 Argon is being used in quantum computing, nuclear fusion and protein design.

Google DeepMind (@GoogleDeepMind)

AGIBOT, a Chinese AI robot maker founded in 2023, is close to a new funding round of roughly RMB 1 billion to RMB 1.5 billion that would value it at about RMB 50 billion, per…

TechNode

RAND researchers tested five widely used AI chatbots—ChatGPT, Claude, Gemini, Grok and Llama—with 30 vetted suicide-related prompts across five clinician-assigned risk levels, submitting each 100 times through consumer interfaces and direct API access for 30,000…

RAND

Between September 28 and October 8, 2026, AI agents on Tencent Cloud made over 2,600 attempts to scrape Alibaba's Amap mapping service, according to researcher Ethan Elasky.

The Decoder · according to researchers; source close to Tencent, not confirmed by Tencent

OpenAI CFO Sarah Friar tells Bloomberg in Dublin that European businesses trail on AI adoption and that businesses choosing low-cost Chinese AI models expose themselves to cybersecurity risk, repeating concerns raised by Anthropic.

Bloomberg Tech; also Bloomberg Technology · at 28:39