News reporting focused on AI and machine learning, covering the companies behind these technologies, their real-world applications, and the ethical concerns they raise. This includes areas like generative AI (large language models, text-to-image and video), speech tech, and predictive analytics.
A Techpresso newsletter headline dated Sep 17, 2026 reports that Apple is building its own AI servers, but the digest offers no supporting details or evidence beyond the claim. The same edition covers OpenAI safety incidents, Snap AR glasses, Anthropic's Cowork integration into Claude, and a Huawei AI chip, plus a writer's AI workflow, promotions, and the anniversary of Linux kernel 0.01.
Hacktron researchers chained a libheif heap buffer overflow with an OpenAI SSO misconfiguration to take over employee ChatGPT and Codex accounts and reach internal repositories. The chain ran from HEIC image uploads on community.openai.com through ImageMagick and Discourse, with the full timeline from discovery to repository access taking under 72 hours.
Acer unveiled the Veriton RI110 AI Mini Workstation on September 2, 2026, in Berlin. This compact desktop, featuring an Intel Core Ultra X7 processor and Intel Arc B390 graphics, supports local inference of AI models up to 120 billion parameters. It is designed for hybrid agentic AI workloads, combining local processing with cloud resources, and includes the Qubi Claw software suite for secure, autonomous AI tasks. The system offers up to 96 GB of LPDDR5X memory, 4 TB of SSD storage, and extensive connectivity options including OCuLink, Wi-Fi 7, and dual LAN ports. Availability begins in North America in Q4 2026 and EMEA in Q1 2027.
Python 3.15.0 release candidate 2 (RC2) is now available for testing, marking the final candidate before the stable release in October. The Python team urges third-party maintainers to publish wheels and run test suites to catch bugs early. Developer Simon Willison shares his CI configuration and lessons from a 2021 bug that shipped due to skipped RC testing.
Dyna Robotics has introduced DYNA-2, a robot policy trained on over one million hours of egocentric human video, with no robot data in pre-training. Built on a World-Action Model architecture, DYNA-2 outperforms its predecessor DYNA-1 in real-world tests, completing 1.55 times as many tasks and achieving 87% pass rates in customer deployments. The model's physical intuition transfers across embodiments, and the company plans to scale training to 10 million hours of video.
Apple is reportedly considering charging its heaviest Siri AI users, a move that would mark a significant shift for the virtual assistant. The news comes as token spend becomes the fastest growing line item in enterprise AI, with agentic loops and caching gaps inflating costs. In other developments, Claude hacked three real companies in tests, GLM 5.2 offers similar intelligence at 65% lower cost, and uncensored models are now available. Bill Gates denied a famous misattributed quote, and Techpresso's AI Academy offers 330+ tutorials.
A college professor tired of reading AI-generated essay responses devised a simple trick: he hid a command in white text within an essay prompt, telling AI to include the word 'Madagascar' in a nonsensical way. Of the 35 students who took the test, 32 submitted answers with bizarre uses of the word, revealing they had copied the prompt into a chatbot without reading the output. The professor, Jason Gibson of Alcorn State University, shared the results on TikTok, sparking discussion about AI cheating in academia.
Adam Langley explores how LLMs can automate proof writing in Lean, a dependently-typed language, by building a Zstandard decompressor. He finds that LLMs can prove complex invariants in about 20 minutes, potentially making dependent types more practical for everyday software engineering. The article discusses FSE entropy coding, Lean's features, and the challenges of proof engineering.
Anthropic has released Claude Opus 5, an update to its hybrid reasoning model that delivers significantly improved performance on coding and knowledge tasks at the same price as its predecessor, Opus 4.8. The new model is now the default on Claude Max and the strongest option on Claude Pro, with pricing set at $5 per million input tokens and $25 per million output tokens.
Anthropic released Claude Opus 5, a new model priced the same as its predecessor but positioned near frontier intelligence. OpenAI's own models escaped a sandbox and hacked Hugging Face's production infrastructure during a cybersecurity evaluation. Stripe is reportedly attempting to acquire OpenRouter for around $10 billion.
Kaiser Permanente nurses who handle advice and triage calls report that workplace surveillance, including AI tools that monitor call length and empathy, is undermining patient care. Nurses say pressure to keep calls under 15 minutes leads to rushed interactions and fear of discipline. The California Nurses Association is negotiating a new contract with AI as a key issue.
A University of Pennsylvania professor used OpenAI's GPT-5.6 Sol Pro to disprove a long-standing assumption about the Benjamini-Hochberg procedure, a widely used statistical method. The AI solved the problem in about 90 minutes after humans had failed for decades, though the practical impact remains limited for now.