A closer look at what changed, what the evidence supports and what comes next.
69 stories
Meta’s English commercial-terms preview narrows crawling purposes. The Verge reports a clarification, while our Danish response still showed different wording. Text changes do not establish crawler behavior.
Anthropic describes six embedded METR staff and publication rights with redaction limits. CNBC’s correction separates SecureBio’s narrower review from that arrangement.
Replit’s CEO describes a break with Anthropic, but TechCrunch reports a dispute and Replit documentation still lists Claude options. We separate partnership claims from account access and task behavior.
Paysend’s CEO describes selective AI access and month-long trials. Our analysis separates his productivity estimate from evidence about completed work and narrower workflows.
Apple makes nonconsensual intimate imagery explicit in moderation rules and clarifies sanctions for hidden features. Our analysis examines what evidence of an effective reporting process would look like.
Two expanded partnerships point to a shared Nvidia driving stack. Our analysis separates the announced architecture from customer availability and evidence of real-world performance.
Glimpse revisits visual evidence while reasoning. Its launch raises practical questions about benchmark scope, task cost and access; we explain how to evaluate those separately.
Vercel reports rapid Jev adoption within its gateway. Team reach, spending share and return rates measure different things; none alone establishes global market leadership.
The reported Chrome extension update expands tab context and browser actions. Our analysis separates a reading task from an authorized action and explains what remains untested.
Analysis of one internal training incident: why a successful deployment and authorized resource use need separate acceptance checks. This is a retrospective case, not a claim about all agents.
NVIDIA announced cuPotential at SEMICON West. The early-access library targets atomic and molecular simulation; its introduction should be assessed separately from claims about wider factory productivity.
Google DeepMind says it worked with GovTech Singapore on an on-device scam detector fine-tuned on local scam conversations. Android Scam Detection is already listed in Singapore on select devices, but Google has not published independent performance,
Meta launched Muse Wonder and Muse Wonder Full as its first frontier models, but access is split across API, Muse Code and a later subscriber-first Muse app rollout. The launch tests whether users will pay for higher-end agentic AI.