News & analysis, rated.
Breaking AI developments, in-depth guides, real-world case studies, and analysis — each one rated so you know what matters.
AI health models need better training data, STAT report finds
STAT's AI Prognosis newsletter examines what quality health data would look like for training useful medical AI models.
Trump revives Biden AI safety rules after Anthropic model alarms
White House creates working group for pre-release model reviews after Anthropic's Mythos demonstrates concerning cybersecurity exploit capabilities.
Agent workloads consume 15x more tokens, break standard serving
NVIDIA details how agentic systems spawn sub-agents and hit 156K context windows, requiring specialized infrastructure to avoid economic collapse.
NVIDIA packages vehicle AI stack from cloud routing to edge inference
DRIVE AGX platforms now handle 7B+ parameter models locally while orchestrating with cloud agents for complex tasks like trip planning.
Microsoft claims AI users produce work they couldn't a year ago
Company's survey of 20,000 workers finds 58% report creating work previously beyond their capability, while organizational factors drive impact more than individual ones.
Microsoft ships Copilot Cowork mobile apps and third-party plugins
Microsoft expands its task delegation AI beyond desktop with iOS/Android apps and connectors to Power BI, Dynamics 365, and external tools.
Microsoft presents 11 systems papers at NSDI 2026
Research spans AI inference optimization, network protocol testing, and cloud infrastructure with independent benchmarks showing measurable gains.
Hugging Face adds private test data to ASR leaderboard
The open-source speech recognition benchmark now includes hidden datasets from Appen and DataoceanAI to prevent gaming.
Anthropic ships 10 finance agents plus Excel/PowerPoint integration
Ready-to-run templates for pitchbooks, KYC screening, and month-end close, with Claude working directly inside Microsoft Office apps.
OpenAI launches self-serve ChatGPT ads with CPC bidding
Beta Ads Manager opens to US businesses while new cost-per-click pricing and conversion tracking replace impression-only buying.
GPT-5.5 Instant cuts hallucinations 52.5% on high-stakes prompts
OpenAI's default ChatGPT model shows significant accuracy gains in medicine, law, and finance while improving personalization controls.
OpenAI flags GPT-5.5 Instant as first high-risk model in series
The company's latest instant model gets elevated safety treatment for cybersecurity and bio/chemical capabilities, marking a classification shift.