Skip to content
Daylede
Daily AI

Daily AI ·

Anthropic accuses Alibaba of largest known distillation attack on Claude

Anthropic accused Alibaba of running the largest known AI model distillation campaign, using nearly 25,000 fake accounts to extract Claude's capabilities over 29 million exchanges. OpenAI and Broadcom unveiled Jalapeño, a custom LLM inference chip built in nine months that promises superior performance per watt. Google made computer use a native tool in Gemini 3.5 Flash, enabling enterprise agents to see and control screens with built-in safety guardrails.

10 sources2 min read

Anthropic accuses Alibaba of largest Claude distillation attack using 25,000 fake accounts Anthropic told US senators and White House officials that operators linked to Alibaba's Qwen lab used nearly 25,000 fraudulent accounts to extract Claude's capabilities in software engineering and agentic reasoning between April and June, totaling 29 million exchanges. The campaign dwarfs earlier distillation efforts and occurred after the White House flagged the technique as a national security concern. 12

OpenAI and Broadcom unveil Jalapeño, a custom LLM inference chip built in nine months OpenAI's first custom Intelligence Processor, designed from scratch for LLM inference with Broadcom and Celestica, has engineering samples running GPT-5.3-Codex-Spark in the lab. Early testing indicates substantially better performance per watt than current state-of-the-art, and OpenAI plans to deploy the chip at gigawatt scale over multiple generations. 34

Google makes computer use a native tool in Gemini 3.5 Flash for enterprise agents Computer use, previously a separate standalone model, is now a built-in tool within Gemini 3.5 Flash, available via the Gemini API and the renamed Gemini Enterprise Agent Platform. The update allows agents to see and control screens, click, type, and scroll, with targeted adversarial training against prompt injection. 56

Anthropic launches Claude Tag Slack integration, now writes 65% of internal code Claude Tag lets teams tag @Claude in any Slack channel to assign tasks; the AI breaks them down, uses connected tools, and reports back in a thread. Internally, the tool already generates 65 percent of code on Anthropic's product team, and an ambient mode enables proactive follow-ups. 7

OpenAI expands ChatGPT ads to Brazil and Mexico next as it courts brands at Cannes OpenAI debuted at Cannes Lions with chief revenue officer Denise Dresser, signaling a push into advertising. The ChatGPT ad business, currently in seven markets, will roll out in Brazil and Mexico within weeks, according to international marketing head Elke Karskens. 8

Google Gemini desktop app adds Speak to Window voice dictation across applications The new Speak to Window feature lets users hold the Fn key and voice commands to any open application without switching tabs. It launches alongside Magic Pointer, which combines on-screen pointing with contextual voice or text prompts for image editing, product comparisons, and chart analysis. 9

Google launches Gemini for Science with hypothesis generation and computational discovery tools Gemini for Science includes three experimental tools: Hypothesis Generation using Co-Scientist with multi-agent idea tournaments, Computational Discovery built with AlphaEvolve, and a Literature Analysis tool. The initiative aims to accelerate scientific discovery across research fields. 10

Daily AI

The most important AI & large language model developments — sourced and cited every morning.