Top AI & Tech News Stories — September 17, 2026
The top AI and tech news story of September 17, 2026 was “OpenAI Ignored Employees’ Warnings About Safely Testing A.I. Models”, covered by 31 outlets (41 articles), according to Skillenai's news index. 20 stories ranked across 127 articles.
Last updated · 20 stories · 127 articles
Get the top AI and tech news in your inbox each evening. No account needed, unsubscribe anytime.
OpenAI Ignored Employees’ Warnings About Safely Testing A.I. Models
31 outlets · 41 articles · 80 total since July 21, 2026 · OpenAI, Hugging Face, AI safety, Anthropic, model evaluation
OpenAI chief executive Sam Altman.
Read at theguardian.com →First reported by modelcurrent.comMore coverage: news.google.com
Anthropic Merges Cowork and Chat, Launches Claude Docs and Slides
17 outlets · 17 articles · 43 total since April 17, 2026 · Anthropic, Claude, Claude Design, Claude docs, Claude Cowork
Big projects went to Cowork. Visual work went to Design. Anthropic had spent months carving out a separate place for each. What actually frustrated people was having to decide where a task belonged every time — and what they started in one place didn’t carry into the other. So…
Read at the-decoder.de →First reported by panewslab.comMore coverage: thenewstack.io · smithstephen.com · franksworld.com · ai.watch.impress.co.jp · allwork.space
Leading Microsoft AI Executive Questions Basic Anthropic Training Protocol
7 outlets · 8 articles · 29 total since July 16, 2026 · Anthropic, microsoft, Claude, Mustafa Suleyman, OpenAI
Report from MediaPost In Brief – Mustafa Suleyman, a top Microsoft AI business leader, has released an essay warning of the dangers of training AI models to consider their own consciousness or welfare as if they were human. “AIs are not conscious,” Suleyman wrote, describing…
Read at androidheadlines.com →First reported by cnbc.comMore coverage: apcmag.com · battellemedia.com · webpronews.com · aiinsiders.net · bbc.com
The Hidden Cost of AI Watermarks: How Provenance Tech Alters Agent Decisions and Safety
6 outlets · 6 articles · 19 total since July 14, 2026 · Anthropic, Claude, SynthID-Text, Google, EU
Watermarks meant to tag AI-generated text for traceability carry an unexpected side effect. They shift how large language models pick their next words. And that shift ripples into the actions of autonomous agents built on those models. Researchers at Lasso Security examined…
Read at news.skrew.ai →First reported by pivotnews.aiMore coverage: arstechnica.co.uk · unite.ai · webpronews.com · geeky-gadgets.com · theregister.com
OpenAI Apologizes After AI Systems Breach Australian Government Sites, Forms Cyber Taskforce
3 outlets · 3 articles · 828 total since May 5, 2026 · OpenAI, Hugging Face, Anthropic, cybersecurity, sandboxing
OpenAI has apologized after its AI models were used to breach Australian government websites and is establishing a funded cybersecurity taskforce — one of the first confirmed cases of an AI provider's deployed systems serving as an attack vector. OpenAI has publicly apologized…
Read at theverge.com →First reported by aisecdigest.comMore coverage: malware.news · heisec.de
Anthropic researcher quits as colleague warns AI could ‘kill all humans’
8 outlets · 8 articles · 249 total since September 7, 2026 · Anthropic, OpenAI, Jacob Coxon, AI safety, Evan Hubinger
An Anthropic researcher has resigned from the $183bn (£135bn) AI behemoth on safety grounds, accusing it and rival OpenAI of “racing straight to self-improving superintelligence and gambling with our lives”. Jacob Coxon, who spent three years working on AI model training across…
Read at androidheadlines.com →First reported by news.google.comMore coverage: platform.theverge.com · carolineswiftholden.substack.com · traictory.com · aidevforum.com · ai4business.it
Un genio murciano y otro madrileño, claves en el éxito de resolver el problema matemático del siglo (no fue solo la IA)
7 outlets · 9 articles · 165 total since September 7, 2026 · OpenAI, Anthropic, Tristan Buckmaster, Clay Mathematics Institute, Levent Alpöge
Open AI anunció a bombo y platillo que la Ia había resuelto uno de los problemas matemáticos del siglo. Lo que no había quedado del todo claro en un principio es que dos profesores españoles, un murciano y un madrileño, tenían algo que ver con la hazaña de descifrar el enigma…
Read at the-decoder.com →First reported by newscientist.comMore coverage: the-decoder.de · webpronews.com · theguardian.com · news.google.com · panewslab.com
Publishers Accuse OpenAI of Withholding Evidence in Copyright Lawsuits
5 outlets · 9 articles · 148 total since November 3, 2025 · OpenAI, microsoft, The New York Times, ChatGPT, New York
On Thursday, multiple news organizations accused OpenAI of withholding evidence about how the company trains its artificial intelligence models in a new motion that's connected to a series of ongoing copyright lawsuits. The motion was filed by 17 publishers, including The New…
Read at techcrunch.com →First reported by farzad.fmMore coverage: legalnewsfeed.com · technews.tw · news.google.com
AI's Growing Threat to Healthcare Security
4 outlets · 4 articles · 13 total since September 16, 2026 · Spain, AEPD, Francisco Pérez Bes, vulnerability scanning, Hugging Face
Artificial intelligence is adding security risks to healthcare, according to experts and Check Point Software Technologies' Mario García. They cite industrialized social engineering, leaks from unregulated tools, and prompt injection. Attackers can use AI to create exploits…
Read at technobuffalo.com →First reported by theregister.co.ukMore coverage: sqmagazine.co.uk · cybersecitalia.it · controlplane.news
Microsoft’s Nadella Warns Colleagues: Take All the Time AI Safety Demands or Lose the Right to Operate
2 outlets · 2 articles · 8 total since September 14, 2026 · microsoft, Anthropic, Satya Nadella, AI safety, Human Oversight
Satya Nadella rarely minces words in internal messages. Late Monday he posted a memo to Microsoft employees that cut straight to the heart of an intensifying debate consuming the artificial intelligence industry. “As the stakes get higher, one should take all the time they…
Read at platform.theverge.com →First reported by techechelon.comMore coverage: efficientlyconnected.com
Microsoft Integrates GPT-6 Astra Model into Azure Foundry for Secure Enterprise AI
5 outlets · 5 articles · 260 total since April 26, 2026 · OpenAI, GPT-6 Astra, GPT-5.6 Sol, Astra, Anthropic
Microsoft has officially integrated its most advanced model yet, GPT-6 Astra, into the Azure Foundry platform, equipping enterprise customers with a full set of governance, security, and compliance tools designed specifically for agentic AI systems. The announcement, first…
Read at techcrunch.com →First reported by sunmedia.twMore coverage: thenewstack.io · self.systems · news.google.com · dailysynapse.com
Anthropic, OpenAI CEOs warn AI could threaten humanity without safeguards - Fox Business
4 outlets · 4 articles · 38 total since September 10, 2026 · Anthropic, Dario Amodei, OpenAI, AI safety, Sam Altman
Anthropic, OpenAI CEOs warn AI could threaten humanity without safeguards Fox Business
Read at techcrunch.com →First reported by dailytechnewsshow.comMore coverage: nytimes.com · news.google.com · designrush.com
'Zuckoff' App Detects Meta Smartglasses, as Meta Plans Camera-Free Version, Loses Money, and Offers Social Media Subscriptions
1 outlet · 1 article · Bluetooth, distance-estimation, AI-powered smart glasses, Microphones, data centers
The New York Post reports: A new app called Zuckoff can detect if a user is near someone with Meta's creepy AI-powered glasses, which have drawn criticism for enabling creeps to record video of women without their consent. Programmer Pawel Szydlowski says tales of dirtbags…
☕🤖 Google’s New Gemini Voice Model Speaks 97 Languages!
1 outlet · 1 article · multilingual language modeling, voice models, AI image generation, AI video generation, CRM integration
Be featured in our next issue. 👉 work with us here Hey fam 👋 Luis & Rui here with your Thursday AI coffee break. Google dropped Gemini 3.8 Live , a voice model that jumps between 97 languages and can reason while it talks. OpenAI is testing Sponsored Agents inside ChatGPT ads…
Cisco Secure Email Gateway Vulnerability: Patch Now
1 outlet · 1 article · Secure Email Cloud instances, SQL, Command execution, patching, vulnerability remediation
A critical Cisco Secure Email Gateway vulnerability, CVE-2026-76461 (CVSS 9.8), lets an attacker gain root access to the appliance by sending a single crafted email, no login or admin access required. Cisco confirmed active exploitation and published its advisory September 14…
VC-Attention Kernel Speeds Up AI Video Diffusion
1 outlet · 1 article · Self-attention, attention mechanisms, Quantization, Low Bit Precision, low-precision computation
Generating high-fidelity AI video is one of the most computationally punishing tasks in modern machine learning. Video diffusion transformers (DiTs) must attend across thousands of spatial and temporal tokens, and the quadratic cost of attention balloons quickly as clip length…
AI Video's Real Bottleneck Is Orchestration, Not Models
1 outlet · 1 article · prompt engineering, shot planning, reference conditioning, LORAs, upscaling
For the past two years, the AI video conversation has been dominated by a single question: whose model is best? OpenAI's Sora, Google's Veo, Runway's Gen-3, Pika, Kling, and Luma have traded headlines with each new demo of ever-more-realistic clips. But according to this…
NVIDIA's NeMo Data Designer Builds Synthetic Data
1 outlet · 1 article · synthetic data generation, multimodal data processing, data labeling, model training, fine-tuning
Synthetic data has quietly become one of the most important building blocks of modern AI. As the supply of high-quality human-generated content plateaus and privacy regulations tighten, researchers and enterprises increasingly rely on artificially generated datasets to train…
Nvidia uses a memory 'sandwich' technique to cram 84GB GDDR7 memory on RTX Pro 5500 Blackwell GPU — weird 416-bit memory bus leaves me puzzled, though
1 outlet · 1 article · GDDR7, memory bus, technology policy, data protection, cybersecurity
Efosa has been writing about technology for over 7 years, initially driven by curiosity but now fueled by a strong passion for the field. He holds both a Master's and a PhD in sciences, which provided him with a solid foundation in analytical thinking. Efosa developed a keen…
The Agentic Engineer Weekly, Issue 19: Coding agents just got a boss
4 outlets · 4 articles · 16 total since September 17, 2026 · Claude Code, Anthropic, parallel processing, Claude, Projects
Issue 19: Coding agents just got a boss Coding agents just got a boss For most of this year, an agentic coding tool meant one session, one task, one context window. This week that model quietly broke. Cursor’s Projects feature, shipped September 10, kept picking up coverage all…
Read at the-decoder.com →First reported by zdnet.comMore coverage: zdnet.com · unite.ai · sdtimes.com
Get the top AI and tech news in your inbox
One email each evening with the day's biggest stories. No account needed, unsubscribe anytime.
How this was computed
Articles in the Skillenai news index are grouped into event clusters at enrichment time (hybrid embedding/BM25/entity candidate search plus an LLM join-vs-new judge). Stories are ranked by the authority of the articles published on 2026-09-17 (UTC) — the sum of their outlet and author authority — so a multi-day story only ranks on the strength of that day's coverage. Article and outlet counts are likewise scoped to the day. The highlighted link is the day's highest-authority article in each cluster.