Latest news, developments and reporting on Large Language Models.
Anthropic says its Bay Area lab found a CRISPR-like enzyme system in bacteriophage DNA, mostly via Claude, with humans running all lab work.
Vambo AI and Tether unveiled new African-language AI models, including a 1.5B system and offline translation tools.
A developer has released ForensicDbg, a post-mortem debugger for native Windows x64 and x86 crashes, in a Show HN post that positions it between Visual Studio and WinDbg and includes an MCP server for agentic debugging.
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei told the UN Security Council that countries should coordinate on AI risks, one day after President Trump rejected what he called a 'globalist scheme' to control the technology.
Google released Gemini 3.8 Flash TTS and Flash-Lite TTS for prompt-based voice design and high-volume production, while opening Google Vids to free 1080p video generation with its Omni model for personal and Workspace accounts.
Le Monde published an article on Sept. 23, 2026, headlined “Aboard the AI Titanic, panic is setting in before the ship has even hit the iceberg.” The available page did not load its full report, displaying only a browser-related error notice.
Apple's iOS 27 adds natural-language event entry powered by Apple Intelligence, redesigned event views, busy/free status for iCloud calendars, extra-large Calendar widgets, and new Siri capabilities, according to 9to5Mac.
Google is adding Gemini connected apps for work, creativity and lifestyle, including Airtable, Adobe and Peloton.
Anthropic says its AI model Claude autonomously discovered a previously uncharacterized enzyme system in bacteriophages, the first result from its new wet lab and an early test of AI's use in science.
A Bridewell study found privacy policies for 20 popular LLMs average 20 minutes to read and score poorly for readability, raising data-understanding concerns.
OpenAI adds voice-agent tools to ChatGPT mobile, letting paid users draft documents and summarize Slack, plus GPT-6 voice models.
Microsoft joined UK police and industry partners to disrupt EvilTokens, an AI-assisted phishing-as-a-service platform that compromised more than 12,000 inboxes at over 10,000 organizations worldwide.
Sen. Bernie Sanders and Rep. Greg Casar introduced the Ban Artificial Superintelligence Act, which would bar development of artificial superintelligence and pause advanced AI models until a new Department of Artificial Intelligence is created. Violators would face up to 20 years in prison.
GPT-6 Astra, Sol and Luna are now in ChatGPT, but only in Work and Codex, not the Chat model picker, according to TechRadar.
On Sept. 23, 2026, Anthropic launched Claude Opus 5.5 and OpenAI launched GPT-6 Sol and Luna, intensifying an AI price war. Leiphone reported OpenAI’s claimed 50% cut rests partly on eliminating its mid-tier Terra model, while Anthropic and DeepSeek rely more on architecture and cache efficiency.
LinkedIn, YouTube, ZeroDrift and Lookout unveiled AI tools Sept. 23, 2026 to verify users, aid creators and fight scams.
At the 2026 Yunqi Conference, Alibaba Cloud AI partners Xingyifu and Huilaidian presented the Modou Cloud AI application ecosystem and Luban Super AI Agent, which serve hundreds of thousands of merchants and handle over 35 billion daily token calls.
TechRadar reports that Palantir CEO Alex Karp believes OpenAI will never pursue an IPO, citing the huge amount of liability it carries. The article's supplied text contains no direct comments or supporting details from Karp.
TechRadar Pro articles published Sept. 23, 2026, describe how AI is surfacing vulnerabilities faster than security teams can validate and fix them, how FICO is pushing purpose-built generative AI for fraud prevention, and how AI-generated websites need extensive customization.
Banma Intelligent released full-modal end-side model AutoOmni 2.0-23B-A3B and debuted AutoClaw 2.0 at the Yunqi Conference.
DeepSeek's new paper, with Liang Wenfeng listed as an author, details DSec, a system that builds more than 5,000 sandboxes per second for agent training and documents the infrastructure and security challenges involved.
Alibaba has appointed Liu Dayiheng, a former Huawei 'Genius Youth' recruit and longtime Qwen pre-training lead, as head of its Qwen LLM team, succeeding Lin Junyang about six months after his departure. He outlined a push toward real-world agents in a 15-minute keynote at the Yunqi Conference.
Nokia's applied research team has open-sourced AnyJev, a training-free Python library that turns open LLMs into calibrated decision models. It borrows Jev's typed decision interface and adds cyclic-shift and prior corrections, while QbitAI contrasts Jev's instant judgments with Decitron's future simulations.
U.S. President Donald Trump called at the UN General Assembly for artificial intelligence to be renamed "Super Intelligence," hours before OpenAI and Anthropic chiefs were due to debate AI safety at the Security Council and ahead of his meeting with China's Xi Jinping.