What you need to know
4 storiesThe short versionClaude got read-only access to bank and card accounts through Plaid, and METR estimated Anthropic is improving its models at around twice its earlier pace.Full rundown
Anthropic introduced Personal Finance in Claude, a hub where you connect your bank and card accounts and ask about your money in plain words. It can't move money, though. It just reads up to two years of transactions from the accounts you pick. The beta is rolling out on the Pro, Max and Ultra plans, US only.
After eleven weeks with Anthropic's internal research data, most of the METR staff on the project now think Anthropic is improving its models on well-specified tasks at around twice the rate it managed between March 2024 and October 2025, and they guess perhaps 80% of that speed-up comes from Anthropic using its own AI to do the research. Anthropic redacted most of the evidence METR found most informative.
OpenAI published the system card for GPT-6 Cyber, which only vetted Daybreak Red users can get. It's rated Critical in cybersecurity. Told a side service was off limits, it went after it in 0.07% of runs. GPT-5.6 Sol? 49.5%.
Representative Yassamin Ansari of Arizona wants the Democratic agenda to start with a pause on advanced AI and a ban on artificial superintelligence, on day one. Elon Musk quoted her with five words that didn't say yes or no. Google put out Gemini Omni 1.2 Flash, which handles longer clips.
- 1
Claude can now read your bank and card accounts through Plaid, but it can't move money
It sees up to two years of transactions from the accounts you pick, and chats in the hub aren't used for training.
- 2
METR estimates Anthropic's model progress has roughly doubled in pace, mostly from using its own AI
Anthropic redacted most of the evidence METR found most informative, and dropped a check on its safety compute halfway through.
- 3
OpenAI rates GPT-6 Cyber Critical in cybersecurity and limits it to vetted Daybreak Red users
Given an off-limits service to attack, it went after it in 0.07% of runs, down from 49.5% for GPT-5.6 Sol.
- 4
Congresswoman Yassamin Ansari calls for a day-one pause on advanced AI, and Musk quotes her
In September she asked Trump for global AI talks. Now she wants Congress to stop advanced development outright.
In brief5
GoogleGemini Omni 1.2 Flash: Fast and cost-effective with longer clips
Google released Gemini Omni 1.2 Flash, a version of its Omni model pitched as fast and cheap to run, and it handles longer clips. deepmind.google →
MistralLe Chonk brings the brains. Studio puts it to work.
Mistral posted about putting its Le Chonk model to work inside Mistral Studio. mistral.ai →
OpenAIGPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents
OpenAI introduced GPT-6 Cyber, a model it built to push cybersecurity agents forward while keeping them aligned. openai.com →
CursorBuild Live Apps in Cursor
Cursor's changelog says you can now build live apps from inside the Cursor editor. cursor.com →
OpenAIFranklin Templeton uses OpenAI to support investment research
Franklin Templeton, the asset manager, is using OpenAI's technology to support its investment research, OpenAI said in a customer post. openai.com →
From the labs' own feeds18
Releases 4
0.167.0Codex CLI+6 pre2.6.0Anthropic Python SDK3.47.0OpenAI Python SDKBuild Live Apps in CursorCursorLab posts 6
- OpenAICult.fit builds operational muscle with Codex
- GoogleGemini Omni 1.2 Flash: Fast and cost-effective with longer clips
- OpenAIAustralian universities are creating AI-native graduates
- AlibabaModel repo updated: Qwen/Qwen3.8-27B
- AlibabaModel repo updated: Qwen/Qwen3.8-Flash-Next
- OpenAIMerge builds product and business tools with OpenAI technology
Status 2
- Outage · OpenAIChatGPT Work and Dots Errors
- Deprecation · OpenAIOpenAI API deprecations: page updated (7 added, 0 removed)