# AI news, Saturday, 19 September 2026

Compiled by Richard Brooks's AI news desk. Canonical: https://richard-brooks.com/ai-news/2026-09-19/

## Alibaba Cloud adds Qwen3.8-Omni-Flash with 1M multimodal context

Models. Text, image, audio and video input with tool calling and web search. Suited to long-form media analysis and multimodal agents without manual chunking.

Source: Alibaba Cloud, https://www.alibabacloud.com/help/en/model-studio/qwen3-8-omni-flash

## Alibaba ships Qwen3.8 LiveTranslate for 60-language realtime speech

Language. Real-time audio and video translation with about 2.3-second latency, speech output in 29 languages, speaker diarization and voice cloning for meetings and localisation APIs.

Source: Alibaba Cloud, https://docs.qwencloud.com/developer-guides/speech/realtime-translation

## Anthropic redesigns Claude Projects for parallel Claude Code agents

Agents. A coordinator runs cloud threads with shared memory and artifacts. Beta for select Pro and Max users; local tools follow later.

Source: Anthropic, https://claude.com/blog/projects-redesigned

## Anthropic says Claude leads 26 percent of its AI R and D

Agents. New R and D Automation Index also covers agent oversight and safety compute share. Follows Amodei's pace-the-frontier push.

Source: Anthropic, https://www.anthropic.com/institute/measuring-pace-of-ai-development

## OpenAI tests Sponsored Agents in ChatGPT ads

Communication. After an ad click, users can open a labelled chat with a business agent. Ads Manager adds prompt tools, HubSpot and Shopify links, and optional auto-translated copy.

Source: OpenAI, https://openai.com/index/reimagining-advertising-with-ai/

## Researchers used Claude Opus 5 to help breach OpenAI accounts

Security. A bug-bounty team chained a Discourse image bug with an OpenAI SSO flaw to reach employee ChatGPT, Codex and internal GitHub. OpenAI says the issues are fixed.

Source: TechCrunch, https://techcrunch.com/2026/09/18/researchers-used-anthropics-claude-to-hack-into-openai/