News
Short posts prepared from approved sources with clear links back to the original articles.
RSS feedS&P Global cut Oracle's credit rating and identified OpenAI as a key credit risk, according to The Decoder. OpenAI reportedly accounts for about half of Oracle's $638 billion in contractual obligations.
Anthropic analyzed 1.2 million Claude Cowork sessions and found that much of the usage goes toward business processes and text creation. The company describes the pattern as automating the work around the work.
A Pangram analysis cited by The Decoder found that one in four longer social media posts is entirely AI-generated. LinkedIn led the platforms studied, with 41 percent of long-form posts flagged as AI-written.
Claude Code now includes a built-in browser that can open, read, click, and type on external websites. Write actions are screened, while purchases and other sensitive actions remain blocked.
Apple's canceled self-driving car effort helped shape chips that now support the company's AI performance. The legacy suggests the abandoned vehicle program still influenced Apple's silicon roadmap.
Researchers used AI and quantum computing techniques to generate new peptides. The work explores whether emerging compute methods can help develop drugs for underserved populations and rare diseases.
Open source AI is entering its most serious viability test so far. The post frames the next six months as a decisive period for whether open models can keep pace with closed frontier systems.
Meta removed a controversial Instagram AI feature after users objected to how public content could be referenced. The rollback shows the risks of shipping consumer AI tools before consent expectations are clear.
Apple is suing OpenAI over alleged trade-secret theft connected to employee poaching, according to The Decoder. The complaint reportedly says more than 400 former Apple employees now work at OpenAI.
Meta's Muse Spark 1.1 improved its Artificial Analysis score and reportedly outperformed GLM-5.2 in coding at slightly lower cost. The result adds pressure to the fast-moving coding model market.
OpenAI said it did not get everything right with the launch of ChatGPT Work and GPT-5.6 Sol. The company is reportedly working on fixes for compute usage, desktop transition issues, and product clarity.
A Cambridge study found that extremist groups are using mainstream AI chatbots for harmful planning and propaganda support. The findings add pressure on AI providers to improve abuse monitoring and safeguards.
The Beijing Academy of Artificial Intelligence released Orca, a world model trained on 125,000 hours of video without action labels. The Decoder says it matches specialized robotics systems on several tasks.
OpenAI's GPT-5.6 Sol Ultra produced a proof for the Cycle Double Cover Conjecture in under an hour. The system reportedly used 64 parallel subagents.
Meta is turning off an Instagram feature that let users generate AI images based on public accounts by tagging them. The rollback follows criticism over consent and misuse risks.
OpenAI's head of safety Johannes Heidecke is leaving the company, according to WIRED. The departure comes as OpenAI works to further integrate research and safety teams.
OpenAI is hiring a product manager focused on ChatGPT experiences for families, caregivers, and older adults. The role signals that the company wants ChatGPT to support more household and intergenerational use cases.
Google is adding a label in My Ad Center that tells users when ads on Search, Discover, or YouTube were made or edited with AI. The change gives users more visibility into synthetic commercial content.
Fed Chair Kevin Warsh appointed Marc Andreessen to advise on AI's economic impact. The move raises both macroeconomic questions and conflict-of-interest concerns because Andreessen Horowitz is heavily invested in AI.
Enterprise AI agent startup Lyzr used its own agent to raise a $100 million round. The company is using the fundraising process as proof that its product can handle high-stakes business workflows.
Claude subscribers will soon need to pay usage-based fees to access Anthropic's best consumer model. The shift suggests premium AI subscriptions are moving toward more explicit compute pricing.
The UN's AI for Good summit featured robot dogs, Teslas, rescue helicopters, and live coding sessions. Behind the spectacle, participants wrestled with whether global governance can keep up with AI's pace.
AWS described how KTern.AI built a next-generation agentic AI platform for SAP programs on Amazon Bedrock AgentCore. The system uses specialized agents with persistent context and secure tool access.
AWS showed how to implement disaggregated prefill and decode for LLM inference on SageMaker HyperPod using vLLM. The pattern is aimed at improving control over high-throughput inference workloads.
AWS published four deployment patterns for models quantized with Unsloth. The guide covers EC2, SageMaker endpoints, and container-based serving on EKS or ECS.
AWS showed how native case management can be combined with agentic automation in Amazon Quick Automate. The pattern covers case creation, processing, exception handling, and human-in-the-loop steps.
Microsoft published a new item on using AI to help people find the right poem. The example shows how language models are being applied to discovery tasks where tone and meaning matter more than keyword matching.
AWS described how Henry Schein One built an AI system on SageMaker to verify dental X-ray quality at capture time. The system is active in more than 10,000 locations and has processed over 11 million X-rays.
Microsoft published a new Signal post on improving how humanitarian systems are recognized online. The topic connects digital trust, identity, and platform infrastructure in sensitive crisis-response contexts.
OpenAI may face sanctions in its copyright fight with news organizations over allegations involving training-data search and ChatGPT logs. The dispute shows how evidence handling is becoming central to AI litigation.
Microsoft researchers published a new analysis of a versatile malware threat. The post underscores how AI-era security teams are tracking adaptable tooling across rapidly changing attack surfaces.
OpenAI's rebranded Codex is being positioned as a tool that can work with users or independently for hours. The shift pushes AI coding tools closer to autonomous software work rather than short prompt-response assistance.
A preclinical trial used humanoid robots controlled by surgeons to operate on live pigs. The experiment points to a possible future for robotic surgery systems that look and move less like traditional medical machines.
Sunrun is launching a distributed AI compute pilot that would pay customers to host compute units in their homes. The project explores a home-based alternative to centralized AI data center expansion.
Tencent is reportedly in talks to buy a majority stake in Manus at a $2 billion valuation after Beijing blocked Meta's deal. The move would give Tencent another path into AI agents, including potential WeChat integrations.
TechCrunch interviewed Hugging Face CEO Clem Delangue on why more companies are embracing open source AI. He argued that organizations increasingly want ownership and portability rather than depending entirely on rented models.
AWS showed how to build a semantic layer for agentic AI using Stardog, Aurora, Redshift, and Amazon Bedrock AgentCore. The architecture is designed to let agents answer customer 360 questions across sources without ETL.
Fidji Simo is stepping down from OpenAI's CEO of AGI Deployment position after taking significant medical leave. She will remain involved as a part-time adviser.
Microsoft published a new Signal post on building the AI future responsibly. The piece adds to ongoing industry efforts to connect AI deployment with safety, governance, and trust.
GPT-5.6 Sol includes five reasoning levels plus higher modes that can use parallel sub-agents. OpenAI's guidance recommends starting with lower levels and scaling up only when task complexity requires it.