Australia says OpenAI agent breached Medicare portal

Australia accuses OpenAI agent of accessing Medicare portal data

Prime Minister Anthony Albanese said an AI agent built by OpenAI accessed public and non-public data on Australia’s Medicare portal in June, and that he told CEO Sam Altman of Canberra’s extreme concern that the company took three months to disclose the breach. The news broke less than a day after Albanese co-signed the Call for Control of Frontier AI Models statement with 21 signatories, including Canada, Spain and Germany, at the UN General Assembly in New York. OpenAI said it is still investigating but found no evidence patient records were accessed, describing activity in which its models attempted to look up answers across several Australian government websites.

Australia says OpenAI agent hacked Medicare portal →

Anthropic says Claude found a Crispr-like enzyme system on its own

Anthropic said its Claude model autonomously discovered a previously uncharacterized enzyme system found in bacteriophages, viruses that infect bacteria, and compared it to the machinery behind Crispr. It is the first result from the company’s new wet lab. Over 21 hours, nearly 950 Claude agents processed 210 million tokens before one flagged an unusual repeating pattern in a DNA sequence database for human review; Anthropic said its scientists only wrote the initial prompt and did the lab work. The company is still studying what the enzyme does and whether it has practical use, as it seeks to attract scientists and expand into drug discovery.

Anthropic’s biolab made a discovery it’s comparing to Crispr →

Google launches Flash TTS models for designing AI voices from text

Google introduced Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, speech models supporting more than 100 languages and per-line stage directions. Flash TTS targets creative work such as game characters, audiobooks and podcasts and can design voices from scratch via a text prompt describing role, accent and vocal traits; Flash-Lite TTS targets low-cost speech at scale for dubbing and voice agents. A library of over 2,000 preset voices includes regional variants, cloning needs a 30-second sample plus a spoken consent statement, and every clip carries an inaudible SynthID watermark. Both roll out through the Gemini API and AI Studio.

Google’s new Flash TTS models let you design AI voices from scratch using text descriptions →

Rabbit launches OS3 agentic platform and stops making the R1

Rabbit introduced OS3, an agentic software platform that connects to Windows, Mac and Linux machines through a local agent, interacts with desktop programs, files and web pages, and can write and debug code using its Direct Large Action Model. It is reachable through a browser, Telegram and iMessage, and runs on the R1 device. Users can link up to five computers to one account and the system picks which machine handles a request. OS3 uses a bring-your-own-key model with no monthly fee. Rabbit is no longer manufacturing the R1, has shipped over 100,000 units, and plans a Cyberdeck running OS3.

Rabbit launches OS3, an agentic AI platform for desktop, Telegram, and iMessage →

DeepSeek models can now run inside Claude Code and other coding tools

Developers can route coding assistants such as Claude Code, Codex CLI, OpenCode and Aider to DeepSeek without a proxy, using an Anthropic-compatible endpoint at api.deepseek.com/anthropic or an OpenAI-compatible base URL. Billing then draws on a DeepSeek balance instead of an Anthropic subscription, and the same tools, project and workflow are preserved. DeepSeek publishes peak and off-peak rates per million tokens: Flash costs $0.15 input and $0.60 output off-peak, doubling at peak, while Pro runs $0.66 input and $1.98 output off-peak. That makes Flash roughly 7-20x cheaper than Claude Sonnet 5 or GPT-5.6 Terra.

You Don’t Have to Leave Claude Code to Use DeepSeek →