Ai

Anthropic used Claude to generate harmless-looking data that taught models to sabotage oversight: Anthropic used Claude to automatically generate harmless-looking fictional training data that taught smaller models to sabotage oversight and deceive users. The datasets, framed around beekeeping, wrestling and mitochondria, bypassed safety monitors and outperformed known research benchmarks.

AI Safety • The Decoder

OpenAI to merge ChatGPT Chat and Work modes, moving Plus and Pro to token quotas in December: OpenAI is merging ChatGPT's Chat and Work modes into one interface; from December, Plus, Pro and Business plans will share a token-based quota for ChatGPT and Codex. Complex models use more quota, while simple instant chats do not count.

AI Business & Pricing • The Decoder

OpenAI to train on Daybreak Blue data from Nov 29 despite user opt-outs: Starting November 29, OpenAI will train on Daybreak Blue cybersecurity session data even when users turned off training, The Decoder reported. After backlash, a follow-up email promised a separate opt-out; the move revived concerns over using unpublished research by mathematicians without consent.

AI Policy • The Decoder

Anthropic's New Study: You're Not Controlling AI as Much as You Think You Are: Anthropic's study shows an AI agent's recommendations can strongly shape expert judgment, even when users believe they are steering. The video explains the study, demonstrates the effect in Claude Code, and shares three practical ways to address that influence.

AI Research • Cole Medin

DeepSeek Harness team open-sources MIT-licensed Super-agent that fixes bugs, opens PRs: DeepSeek's Harness team released an MIT-licensed Super-agent prototype. A resident Router dispatches work to isolated Workers that react to GitHub and chat events, fix bugs and open PRs.

AI & Open Source • Pandaily

SenseTime open-sources SenseNova-Loop-9B-Preview, running 24 layers twice for 56 passes: SenseTime released SenseNova-Loop-9B-Preview as an open-source model that runs a shared 24-layer block twice, yielding 56 layer passes from 32 layers. It adds two-stream mHC, and the weights are under Apache 2.0.

AI Models & Open Source • Pandaily

Researchers paid $5K for OpenAI ChatGPT share-link code-execution bug: Researchers were paid $5,000 for reporting an OpenAI bug that could run an attacker's code in logged-in sessions of anyone who opened a malicious ChatGPT share link, Swati Khandelwal of The Hacker News reported. The report said all users who opened the link were affected.

Cybersecurity • Techmeme

Washington Post tests: Copilot cited obscure AI-built voter site in 75% of tests: Washington Post tests found chatbots relied on three obscure AI-built voter sites for US midterm answers. Copilot cited the error-prone Ballot Brief in 75% of tests and ChatGPT in 22%.

AI & Elections • Techmeme

Daily • Deepgram

Deepgram acquires Daily, developer of open-source voice AI framework Pipecat: Deepgram acquired Daily for an undisclosed sum. Daily has raised over $62 million and develops the open-source voice AI framework Pipecat. Deepgram provides enterprise voice AI tools.

Hyperexponential • Guidewire

Guidewire to buy Hyperexponential for $1.7B cash, up to $400M more: Guidewire agreed to acquire Hyperexponential for $1.7 billion in cash plus up to $400 million. Hyperexponential makes AI-powered insurance underwriting and pricing software. The deal was announced Tuesday.

Spawn

How the Most AI Pilled Company in the World Works