Tencent’s real-time Hy-ASR-3.0 preview accepts up to one minute of 16 kHz mono PCM, while external context input, hotwords, diarization and voice-activity detection remain outside the current API contract. Synthetic narration produced by Tachibana Navi.

OpenAI4S now presents 34 research skills on its main branch, while its only tagged release predates later workflows and fixes. Scientists need to label the exact commit or package they test. Synthetic narration produced by Tachibana Navi.

IBM found a $630,000 average cost gap for AI-related incidents in a sample of 602 already-breached organizations. Within the AI-related subset, 92% lacked proper access controls—but the observational benchmark does not show that AI caused the higher costs. Synthetic narration produced by Tachibana Navi.

Microsoft’s source-only Skill Recorder turns one desktop demonstration into an editable agent procedure, but cloud analysis can receive screenshots, window titles, browser URLs, clipboard previews and narration. Synthetic narration produced by Tachibana Navi.

Police in Japan handled more cases involving sexual deepfakes of under-18s in six months than during all of 2025. Most involved junior-high or high-school students, and reported classifications put 83 cases inside school peer networks. Synthetic narration produced by Tachibana Navi.

Separate research teams used the same proprietary model to pursue one unclonable-encryption problem, producing closely aligned preprints within hours. The result makes model disclosure, human verification and priority harder to separate. Synthetic narration produced by Tachibana Navi.

Alibaba has opened Qwen3.8-Max through QwenCloud with a one-million-token context window, while promising downloadable weights next week. Developers can test the hosted model now; licensing, artifacts and hardware requirements for the promised checkpoint remain unanswered. Synthetic narration produced by Tachibana Navi.

INTERPOL says AI appeared across the attack chain in cases observed by 36 respondents to its 49-country African survey, while incomplete loss and victim data complicate the response. Synthetic narration produced by Tachibana Navi.

MiniMax released a 498.3 GB H3 repository with local 768p generation, while multimodal orchestration and 2K finishing remain hosted. The custom license also sets an exact commercial-revenue trigger for prior authorization. Synthetic narration produced by Tachibana Navi.

Seedance 2.5 is rolling out with 30-second joint audio-video generation and a 50-file reference budget, while ByteDance’s launch page gives developers no date, price or operating details for ModelArk API access. Synthetic narration produced by Tachibana Navi.