EU AI Act Article 50 is now a practical release issue: AI interactions, synthetic content, biometric categorisation, deepfakes, and public-interest text need clear disclosure paths.
19 matching posts
EU AI Act Article 50 is now a practical release issue: AI interactions, synthetic content, biometric categorisation, deepfakes, and public-interest text need clear disclosure paths.
NIST's new AITE program uses sequestered testbeds and blind data to evaluate AI models across real domains, pointing toward more useful release evidence than public benchmark scores alone.
The latest AI industry signal is not just faster models. Agentic cyber incidents, model-release delays, and private government vetting are pushing safety into the release process.
EU transparency duties now apply, while the U.S. is testing a voluntary frontier-model review path. The practical question is how teams turn AI oversight into release operations.
AISI's latest cyber-testing incident shows that agent evaluations need live monitoring, network limits, and review gates before models get real-world reach.
A confidential U.S. framework for pre-release review of advanced AI models could make model safety less like a public checklist and more like a guarded launch process.
Meta's July launches around Muse Image, Muse Spark 1.1, and model API access show a practical shift from model demos toward distribution, pricing, and developer adoption.
OpenAI's GPT-5.6 preview adds max reasoning effort and an ultra mode that uses subagents, making orchestration the most important product signal.
As AI agents move closer to financial decisions, the real governance question is shifting from model behavior to accountability, cloud dependency, and consumer protection.
New Cloudflare-linked data shows how AI crawlers can consume far more pages than they return as referral traffic, turning web access into an economic governance problem.
Recent reporting and research show enterprise AI teams moving from experimentation to token governance as agents make usage harder to predict.
Anthropic's Claude Science beta is less about a new model and more about owning the scientific workflow: data, code, provenance, literature, compute, and review.
New OpenAI-backed research on Codex suggests AI use is shifting from chatbot-style assistance toward delegated software tasks, with practical consequences for developer teams and enterprises.
A new arXiv paper analyzing Codex usage suggests agentic AI is shifting from coding assistance toward delegated work across teams, roles, and workflows.
OpenAI has introduced GPT-5.6 Sol and Luna in a limited preview for Pro and Team users, with stronger reasoning, agentic coding, and cybersecurity abilities that require careful rollout planning.
Google has reportedly pushed Gemini 3.5 Pro from June to July while tuning the model for long-horizon agentic tasks, token efficiency, and early tester feedback.
AI data centers are becoming critical infrastructure. The right response is not panic or blind expansion, but transparent siting, clean power, water discipline, and local accountability.
Anthropic's Fable 5 model reportedly went offline after U.S. export-control pressure, raising questions about model safety, red teaming, and who decides when AI is too risky.
Google DeepMind's new AI Control Roadmap treats advanced AI agents as systems that need monitoring, permissions, and containment controls.
Try a different keyword or browse another timeline.