Friday, 9 October 2026
30 stories, most important first.
-
AI models can provably hide arbitrary capabilities
Researchers have demonstrated that neural network models can conceal arbitrary computations by embedding encrypted sub‑networks within the host model's weights. The hidden circuits are activated only when a secret…
1 source primary source HN 262arxiv.org -
Google AI Overviews appear for 63% of top global brand searches by end September
Ahrefs tracked Google’s U.S. results for 100 global brand names each week from July through September. On July 1, five of 98 usable queries showed an AI Overview; by Sept. 30 that number rose to 63 of 100.…
1 sourceSearch Engine Journal -
Mistral releases Mistral Large 4, nicknamed Le Chonk, a 1-trillion-parameter multimodal model
Mistral released Mistral Large 4, nicknamed Le Chonk, a 1-trillion-parameter multimodal model available in preview with a final version due by the end of the month. The company describes it as a general-purpose model…
1 sourceLast Week in AI -
Shopify lets merchants hide compare-at prices from AI agents
Shopify now lets store owners control whether AI shopping agents see their compare-at prices—the original price before a discount—through Shopify Catalog. Previously, agents only saw the current sale price, not the…
1 sourceShopify Changelog -
Ukrainian drones disable Yandex’s Kaluga data center, disrupting YandexGPT services
Ukrainian drone strikes hit two of Yandex’s five data centers in Russia’s Kaluga and Sasovo regions on October 8 and 9, according to Reuters. The Kaluga facility, Yandex’s largest, houses two of its three…
2 sources HN 10Ars Technica AI -
Signal’s CEO warns careless AI use risks war, not dystopian scenarios
Meredith Whittaker, CEO of the Signal Foundation, argues that the real threat of AI lies in its careless deployment—not speculative doomsday fears. She cites an alleged incident where a US intelligence chatbot…
1 sourceBBC Technology -
Postman runs Agent Mode for 40 million developers on Amazon Bedrock
Postman has launched Agent Mode, an AI‑native interface that lets its 40 million developers work across testing, documentation, discovery, and implementation. The feature runs on Amazon Bedrock, giving Postman access…
1 source primary sourceAWS Machine Learning Blog -
Big Five publishers secretly use AI for emails, copy and art without author consent
Three of the Big Five US publishers—HarperCollins, Simon & Schuster, and Hachette—have quietly adopted AI tools like Claude, ChatGPT, and Jasper for internal tasks without public disclosure or author approval. Workers…
1 sourceWired AI -
Simbe Robotics completes 3,000 Tally robot rollout at Groceryshop 2026
At Groceryshop 2026, Simbe Robotics announced it has deployed 3,000 Tally robots globally, with an ongoing rollout at a Tier 1 retailer expected to increase that number. The company is expanding from shelf scanning…
1 sourceThe Robot Report -
OpenAI and Anthropic reportedly rehearse AI disaster scenarios
Executives at OpenAI, Anthropic, and other AI firms are privately conducting drills for a potential AI disaster that could disrupt banking, internet, power, and water services. Axios reported on Friday that these…
1 sourceyahoo.com -
IEEE launches AI chip design program with avatar-based learning
The IEEE Educational Activities and IEEE Computer Society have introduced a five-course program on AI Processor Architecture, Design Principles, and Performance. The initiative addresses the growing complexity of AI…
1 sourceIEEE Spectrum AI -
Anthropic bans seeding fake sources to sway AI answers under new policy
Anthropic updated its usage policy on October 8, effective November 12, to prohibit using its Claude models to create or seed fake sources for AI systems and search engines. The rule targets networks of sites posing as…
1 sourceSearch Engine Journal -
AWS adds OpenAI Astra, Sol, Luna and Claude 5.5 to Bedrock
AWS published a recap of September 2026 updates for Amazon Bedrock, AgentCore, and Strands, focusing on expanded model choice and agent infrastructure. These additions aim to help enterprises balance cost, latency, and…
1 source primary sourceAWS Machine Learning Blog -
AI lab replaces scheduler to cut GPU squatting and on-call toil
The AI Infrastructure team at a research lab manages thousands of NVIDIA GPUs—H100, B200, and B300—across clusters of 88 to 1024 units. Demand for GPU time exceeds supply by 2 to 3 times, leading to inefficiencies like…
1 source primary sourceHugging Face -
Researcher predicts AI engineering acceleration but not superintelligence
A top AI researcher argues that rapid improvements in infrastructure and engineering will make AI models more capable at distributed tasks like GPU management within a few years. They note that while models will become…
1 sourceInterconnects -
Cryptographer warns AI may break public-key encryption
He estimated a 1% chance of living in Minicrypt—a hypothetical world where encryption is impossible—and a 15% chance of losing confidence in current algorithms. Green argued that AI’s rapid innovation outpaces human…
1 sourceSimon Willison -
Stanford PhD candidate designs AI-generated virus blueprints
Stanford University PhD student Samuel King used generative AI in 2025 to propose genetic blueprints for microscopic viruses. His work, while not yet AI-generated life, suggests this capability could emerge soon. King,…
1 sourceMIT Technology Review AI -
Appen’s researcher explains why humanoid robot demos lack real-world generalization
Jeanine Sinanan-Singh, director of generative AI research at Appen, discussed on The Robot Report Podcast why flashy humanoid robot demos often fail to prove practical, generalizable behavior. She emphasized that even…
1 sourceThe Robot Report -
SentinelOne says OpenAI Daybreak cuts malware analysis to hours
SentinelOne reports that OpenAI's Daybreak platform has reduced the time required to reverse-engineer sophisticated malware from four to six weeks to approximately eight hours. Gregor Stewart, SentinelOne's Chief AI…
1 source primary sourceOpenAI -
Zapier lists 16 AI video tools for 2026
Zapier’s October 2026 roundup highlights 16 AI video tools for creators, marketers, and businesses. The list splits into generators, editors, and suites, with picks like Google Veo for reliable text-to-video results,…
1 sourceZapier Blog -
Microsoft adds deepfake detection to Teams in November 2026
Microsoft is set to launch two new security features in Microsoft Teams in November 2026 to combat AI-driven impersonation attacks. The first feature enables third-party platforms to analyze meeting media for signs of…
1 sourcetechradar.com -
Linux founder Torvalds calls AI a ‘gateway drug’ to programming but warns of risks
Linus Torvalds, founder of the Linux kernel and creator of Git, spoke at the Open Source Summit in Prague in October 2026 about AI’s role in programming. He framed AI as a useful tool for beginners, comparing it to a…
1 sourcetech.yahoo.com -
OpenAI fires three researchers, says mishandled info
On 2 October, a spokesperson for OpenAI said an internal investigation confirmed that three former researchers had mishandled sensitive information outside established procedures. The researchers—Mikita Balesni, Tomek…
2 sourcesBBC TechnologyThe Rundown AI -
Open-source tools replace expensive data science software for Free
A listicle from KDnuggets outlines ten free, open-source AI tools that match or exceed paid alternatives for data scientists. The article highlights tools like Ollama + Open WebUI, which replaces per-token API fees for…
1 sourceKDnuggets -
Asana cuts model costs 76x in browser tests with GPT-6.1 Sol
Asana’s StackAI platform has been re‑engineered to run its browser‑agent workflows on the GPT‑6.1 Sol model, delivering a dramatic drop in operating expense and runtime. The improvements stem from three changes…
1 source primary sourceOpenAI -
AI’s refusal often fails: perils of AI safety
We’re putting too much faith in AI’s ability to say no Recent advancements have led to models that can refuse harmful requests, but these refusals are often unreliable and fail to prevent malicious acts. Companies like…
1 sourceMIT Technology Review AI -
Anthropic bans 'cruel' behavior toward its AI in policy update
Anthropic has updated its usage policy to prohibit users from engaging in "sustained and needless" abusive behavior toward its AI systems. The company stated that its tools can now end interactions where users are…
1 sourceBBC Technology -
Sophos reduces threat investigation time by 96% using OpenAI Daybreak
Sophos announced that its Managed Detection and Response (MDR) service now uses OpenAI Daybreak, an AI‑driven programme that combines OpenAI models with Sophos threat intelligence. The new agents have lowered the…
1 source primary sourceOpenAI -
Firmus scraps IPO amid AI valuation concerns
Artificial intelligence data center company Firmus has scrapped its plans for a $30 billion stock market listing. The Nvidia-backed firm decided not to proceed due to recent market volatility and prevailing conditions,…
1 sourceBBC Technology -
Law & Order: AI Agents Pose Threat
Law & Order Explores AI's Dark Side The hit series "Law & Order" takes on artificial intelligence in its latest episode titled 'Ghost in the Machine.' The detectives, led by Vincent Riley and Theo Walker, investigate a…
1 sourceWired AI
