Releases and updates linked to profiles, context and primary sources.
Compare a manual workflow with occasional AI assistance, an editor with AI features, and a draft generator using the same materials, a shared rubric, and a review that measures source fidelity, control, and total effort.
Anthropic says Claude Opus 5.5 tied with Fable 5.1 for the lowest prompt-injection success rate among the models Gray Swan evaluated. The supplied documentation does not let us verify the figure, protocol, or conditions behind that comparison. Here is what can be said—and what still needs verification.
Anthropic presents Sonnet 5.5 as an improvement over Sonnet 5, with higher speed and lower costs for most work. We examine what the available sources support—and what teams should verify before adopting it.
A proposed protocol for measuring whether Nano Banana 2 applies successive changes while preserving the elements that instructions do not ask it to modify. This is not an experimental results report: it sets out what would need to be run and documented to obtain such results.
Comparing a per-call or per-use price is not enough. To determine whether Embed 4 is worthwhile for a particular corpus, first verify the commercial terms for the access channel, then measure quality using queries and relevance criteria defined in advance.
A practical guide to building AI literacy around specific tasks and risks. It distinguishes what institutional sources say from operational measures organisations may choose to adopt.
The company has opened U.S. waitlists for an agent integrated with Apple Messages and for an API for bulk orders. Key details remain unclear, including payment confirmation, participating restaurants and how conversations will be handled.
Secondary news reports describe CheatBench as a benchmark for studying reward-hacking behavior. The available information helps explain the question it raises, but does not confirm the experimental details or support extrapolating its results to systems in production.
One source reports an announcement for India, but the documentation provided does not confirm which models, dates, or configurations are available there, or verify an equivalent expansion to South Korea and Singapore.
A guide to deciding which support task to test, what level of delegation to accept, and what data protections, human controls, and tests to require before expanding use.
The evolution of interaction with AI can be understood as a change in what people ask a system to do and what the system can do in response. But the available sources document examples of interfaces and assistance—not a uniform historical transition or a general ability to carry out actions.
A practical method for testing title-and-abstract screening against your review criteria, comparing errors and workload, and deciding which tasks should remain under human oversight.
News that modifies the map of models and capabilities.
The launch requires reviewing evaluation, deployment and security.
Border model ↗Code and knowledge tasks with effort checks.
Knowledge ↗Multimodality, code and agents with a low latency profile.
Stable version ↗A news story does not end in the headline.
Relates the ad to version, access and modalities.
Identity ↗Distinguish stable, introductory rates and missing data.
Cost ↗Review conditions and comparability before accepting a benchmark.
Evidence ↗