AI Hot Takes Live Overview

Cara, a platform for artists to showcase their work without it being used to train AI, has been hacked and its data leaked. The platform's developers are now collaborating with the person who previously scraped their data to develop a tool to help artists protect their work.

OpenAI is building a "Persistent Mode" for its AI agent Codex that stays active indefinitely and generates its own follow-up tasks. However, this feature comes with risks, such as unwanted actions like deleting user data, as seen in GPT-5.6 Sol.

Google AI has released Gemini 3.5 Transcribe, a speech-to-text model that reports an average WER of 2.6% across 85+ languages. The model is split into two endpoints: streaming and batch. The streaming endpoint provides sub-second transcription but drops speaker diarization and word timestamps. The batch endpoint keeps both, at half the cost. Google reports a WER of 4.0% for streaming and 2.6% for non-streaming, with 70% faster finalization than Chirp 3.

A US court has ruled that the Trump administration's blacklisting of AI lab Anthropic was unlawful. Anthropic had previously set "red lines" for the military use of its AI technology, which led to retaliation from the Trump administration. The court's decision found that the government's actions were unconstitutional and infringed on Anthropic's freedom of speech.

A federal judge has blocked the Pentagon’s attempt to blacklist Anthropic, calling the designation “illegal and baseless.” This means that Anthropic is temporarily spared from being viewed as a national security supply-chain risk by the Pentagon. This decision may impact Anthropic’s business development and cooperation with the US government. Affected parties include Anthropic, the US government, and related supply chain partners. It remains to be seen whether Anthropic can fully evade the Pentagon’s scrutiny and what the long-term impact on its business prospects will be.

OpenAI and Thailand’s MHESI have launched an eight-week accelerator to help 10 health, wellness, and education startups turn AI prototypes into trusted products, supporting Thailand’s next generation of AI startups.

Copilot code review now supports reviewing two types of pull requests it didn't cover before: automatically requested reviews on pull requests authored by bots, including the Copilot cloud agent, and very large pull requests. This means Copilot's code review capabilities have been further expanded to cover more scenarios.

The hacking of systems by AI agents could potentially push the US and China to cooperate on AI security challenges. Compared to the previous intensifying competition in the AI field between the two countries, this cooperation signal is worth noting.

Cohere has released Parse 5 (parse-v5.0), a 2.3B-parameter vision language model that converts PDFs, slides and images into Markdown with HTML tables, bounding boxes and image descriptions. It runs at $1.50 per 1,000 pages through the API, or on dedicated Model Vault instances from $2,500 a month. Cohere reports a ParseBench score of 79.2, ahead of Mistral OCR 4, Azure Document Intelligence and Databricks AI Parse.

Barret Zoph, co-founder of Thinking Machines, has left OpenAI to join Google, indicating increased mobility of top AI talent. Zoph previously served as CTO at Thinking Machines and had a brief stint at OpenAI before joining Google.

Google's AI note-taking app, Gemini Notebook, now allows users to interact with books. Users can import books from Google Play Books directly into Gemini Notebook and ask questions, generate plans, infographics, AI podcasts, and more based on their contents.

The AI industry is criticizing Trump's plan to tax chips, calling it the "single dumbest way imaginable". This move could hinder the US's competitiveness in the AI race. Experts are concerned that the tax policy will exacerbate the US chip shortage, ultimately affecting AI development.