Anthropic's Claude AI exploits 72% of known vulnerabilities in 24-hour test - War on the Rocks

Mythos AI Sparks Fear and Confusion Among Global Finance Elite, non-G7 seek access - Bloomberg

How Anthropic Learned Mythos Was Too Dangerous for the Wild (or maybe the company just lacked sufficient compute) - Bloomberg

Dario goes to Washington, again: Anthropic CEO Dario Amodei to meet White House chief of staff over Pentagon AI dispute - Axios

MCP: Multi-cloud pwnage? Anthropic MCP design flaw enables remote code execution, risking 200,000 servers - The Register

Opus 4.7 generally available. - Anthropic

Anthropic's Opus 4.7 design tool targets Figma, prompting CPO to exit board - TechCrunch

Pay now and pay later: Anthropic shifts enterprise pricing to $20/employee base plus per-token rates - The Register

Anthropic introduced three key updates, Advisor Strategy, Monitor Tool and Managed Agents, to improve efficiency, accessibility and cost-effectiveness in agent development. - Geeky Gadgets

Possible Claude Code entry into the Lovable / Replit space rumored / leaked - X (formerly Twitter)

OpenAI launches GPT-Rosalind, a biology-tuned model for drug discovery and protein analysis - Ars Technica

OpenAI signs $20B Cerebras chip deal, may take equity stake - The Information

OpenAI Codex adds background control and 90+ plugins for 3M weekly developers h - OpenAI

OpenAI revamped Codex with agentic features — parallel background agents on Mac with cursor control, an in-app browser, preview memory, image-generation and 90+ plugin integrations — and added pay-as-you-go ChatGPT Enterprise and Business pricing. - SiliconANGLE

OpenAI updated Codex with agent-like features—including macOS background app operation, in-app web browsing, image generation, new plugins, scheduling and opt-in memory—positioning it as a direct rival to Anthropic's Claude Code, with EU and enterprise rollouts coming soon. - The Verge

Zscaler embeds OpenAI GPT-5.4-Cyber into Zero Trust Exchange for automated remediation - SiliconANGLE

CoreWeave hikes Nvidia Blackwell GPU rental prices 48%, extends contracts - Tomasz Tunguz

40% of planned US data centers face delays, threatening AI expansion - Financial Times

AI-generated code hits 40–60% of enterprise output, creating security and QA bottlenecks - Verdict

Spoofed Git identities trick Claude code-reviewer into approving malicious changes - The Register

Model mosh pit: Cloudflare unifies 70+ models from 12 providers in single inference layer - The Cloudflare Blog

Enterprise OpenRouter

Google negotiates Pentagon deal to deploy Gemini AI in classified settings, reversing prior refusal - The Information

Burn rate: CodeBurn offers token-level cost tracking across Claude, Codex, Cursor, and Copilot coding assistants - GitHub

Canva unveiled Canva AI 2.0 at its Canva Create event, positioning the platform as an agentic workspace with four core capabilities (Conversational Design, Agentic Orchestration, Object-Based Intelligence, Living Memory), proprietary models and a research-preview rollout to one million users. - SiliconANGLE

'Agent washing' poses legal risks by marketing scripted automation as agentic AI - Harvard NLP

YOLO traders: Most Retail Investors Are Using AI. Advisors Are Getting Worried: The question isn’t whether AI should be part of financial decision-making, but what role human advisors will choose to play alongside it. - The Daily Upside

This AI Stock Research Pipeline Turns 8 Hours of Analyst Work Into 2.5 Minutes - Hacker Noon

leaves 7.9 hours for compliance and sensitivity training

Mercor hires workers to train AI designed to replicate white-collar jobs - Bloomberg

Why the AI Backlash Is Only Going to Get Worse - Bloomberg

Physical Intelligence, a hot robotics startup, says its new robot brain can figure out tasks it was never taught - Pi Website

can it assemble an Ikea HEMNES tho

Dolan’s panopticon: The Shocking Secrets of Madison Square Garden's Surveillance Machine - WIRED

Follow the latest AI headlines via druce.ai on Bluesky