AI is moving from chat windows into decisions, devices, and daily work

OrcaRouter’s gated OrcaCyber Zero 1.5 is reported to support a one-million-token context window, with strong results on two cybersecurity benchmarks. For security teams, that could mean examining more material in one pass—but benchmark scores don’t tell us on their own how it performs in real investigations.

Researchers at the Icahn School of Medicine studied whether safety prompts can help AI make safer decisions in clinical tasks. If the approach holds up beyond the study, it could offer builders a practical layer to test alongside deeper clinical safeguards—not a replacement for expert oversight.

A report describes Microsoft’s Decision-1 as a model for ranking options and guiding agent decisions, rather than primarily generating prose. That distinction matters for teams building agents: reliable choices and clear ways to evaluate them may be more useful than another fluent answer.

Qualcomm says some AI companies have asked for phones capable of running models with at least 100 billion parameters by 2028. It’s a striking signal of interest in on-device AI, though a request is not a shipping product—and practical questions around memory, power, and cost remain.

The developer behind ArtCraft has announced open-source projects modeled on spreadsheets, presentations, and word-processing tools. For users, the appeal is more choice and potentially lower costs; for builders, the hard part will be delivering compatibility and a dependable everyday experience.

A Medium article says GPT-6 is now ChatGPT’s default and focuses on a shift toward tools and actions. Treat that as a claim to verify, rather than a product update confirmed by the other sources here; either way, interfaces that hand work to tools make transparency and user control especially important.

Indiana Governor Mike Braun has announced a new AI-powered tool as part of a government-efficiency effort, according to the local report. Public-sector deployments are a useful test of whether AI can improve services while remaining accountable to the people affected by its decisions.

Forbes’ business-tech roundup includes a list of AI website builders. For small businesses and independent creators, these tools may lower the barrier to getting online—but it’s worth comparing how much control users retain over design, content, and portability.

Google Labs’ new Playground service lets users create video games with AI, according to the report. It could make experimentation more accessible to people without game-development experience; the interesting question is how much creative control users have once the first draft is generated.

True Fit’s new Fit Intelligence Layer aims to use data to improve AI sizing and reduce returns for fashion retailers. If it works, shoppers could get more useful size guidance and retailers fewer avoidable returns—but the quality of the recommendations will depend on how well the underlying fit data reflects real customers.