A closer look at what changed, what the evidence supports and what comes next.
69 stories
OpenAI says frontier-model training now runs without internet access. The announcement concerns internal research environments and needs to be read separately from customer-product permissions.
Etsy prohibits AI-generated functional instructions while continuing to permit seller-prompted artwork under conditions. Disclosure, human participation and product category are separate checks.
METR’s public overview describes a historical OpenAI monitoring pilot. The findings, later system changes and OpenAI’s unevaluated patch claims support different conclusions and should remain separate.
Canva Craft reportedly plans a design before rendering it. A useful production test should distinguish visual appeal from correct text, constrained edits and accepted final assets.
Nvidia’s Minerva announcement connects infrastructure and workload monitoring. Our analysis separates fewer alerts from accurate diagnosis and verified workload recovery; the announcement supplies no independent comparison.
1Password says its agent sign-in flow keeps credentials outside the model. The remaining question is what an authenticated agent is allowed to do, and how session boundaries are explained.
Flint’s model card separates total parameters, active parameters and checkpoint formats. Those distinctions matter when assessing laptop compatibility and the evidence behind AI-assisted coding claims.
Claude Duet’s reported periodic account checks distinguish proactive briefings from continuous monitoring. The preview raises practical questions about alert timing, notification judgment and shared usage limits.
IBM reports revenue growth while management describes AI’s share of consulting work. Those measures do not isolate AI’s incremental revenue or profitability; the distinction matters when comparing enterprise suppliers.
Gemini is reaching Google Maps on CarPlay. The reports describe conversational navigation, but do not establish reduced distraction, complete regional availability or broader personal-data access.
Nvidia projects a shorter deployment at a prepared site. Our analysis explains why buyers should define the start and finish of that clock before comparing offers; the announcement is not an independently verified installation.
Newly reported Muse estimates include seven- and fourteen-day retention. Downloads, daily activity and cohort return still describe different populations and time windows.
Electric Plant describes an AI model for plant electrical signals. Our analysis distinguishes recording a signal, inferring a condition and improving a growing decision.