
AI agents take over your login, your work, and apparently your hacking too
From 1Password letting agents log in for you to Chinese hackers using agents to breach thousands of companies — agents are having a very big day.
Subscribe free →
DeepSeek Raises $15B Days After the US Tried to Banish It
Just days after the U.S. ordered federal agencies to rip out DeepSeek's technology over national security concerns, the Chinese AI lab reportedly closed one of the largest funding rounds in Chinese history at $15 billion. The timing is either defiant or coincidental — either way, the message is clear: US restrictions aren't slowing DeepSeek's momentum at home.
Why it matters: A $15B war chest means DeepSeek can continue competing with US frontier labs on model quality and price, US bans notwithstanding.

Chinese Hackers Used AI Agents to Breach Thousands of Companies Through ConnectWise
Microsoft has attributed a massive supply-chain attack on IT management platform ConnectWise to a China-aligned hacker group — and the scary part is how they pulled it off: AI agents were used to automate the lateral movement into thousands of ConnectWise customer networks at a speed no human team could match. This is one of the clearest real-world examples yet of AI being weaponized for large-scale cyberattacks.
Why it matters: Offensive AI agents can now scale a single breach into thousands of compromised organizations in the time it used to take to hit a handful — the attack surface calculus just changed permanently.

1Password Lets AI Agents Log Into Your Accounts Without Ever Seeing Your Password
1Password's new feature creates a credential layer specifically for AI agents — letting them authenticate into your accounts using tokenized credentials rather than your actual passwords. The agent gets access; your secrets stay locked. It's a smart solution to a problem that was going to get very messy very fast as agentic workflows touch more personal services.
Why it matters: As AI agents become de facto digital employees handling real tasks, secure credential delegation isn't a nice-to-have — it's the infrastructure that makes agentic computing safe enough to actually deploy.

Anthropic Launches Claude Duet, Its Bet on Persistent AI Agents
Anthropic is rolling out Claude Duet — a new app experience built around long-running, agentic Claude interactions — to Pro and Max plan users via waitlist. Details are still thin, but the positioning is explicit: this is Anthropic's answer to the question of what AI looks like when it moves beyond single-turn chat into sustained, autonomous work.
Why it matters: Every major AI lab is now shipping an agent product — Duet puts Anthropic directly in the ring with OpenAI's operator-style agents and Google's Gemini agentic suite, and the differentiation battle starts now.

Microsoft's AI Beats Doctors on Hard Diagnostic Cases — With an Asterisk
A study published in *Science* found Microsoft's AI diagnostic system outperformed generalist physicians on difficult medical cases. Mustafa Suleyman is using the result to beat the drum for "medical superintelligence." The catch: TechCrunch notes that most of the performance lift appears to come from the underlying foundation models, not Microsoft's system-level orchestration on top.
Why it matters: The headline numbers are real, but if the gains are model-driven rather than system-driven, Microsoft's diagnostic layer is less defensible than the press release implies — and the underlying model race matters more than ever.

Uber Rolls ChatGPT Work Out to All 30,000 Employees
Uber has gone company-wide with ChatGPT Work, OpenAI's enterprise agent platform, deploying it to its entire workforce after a smaller pilot. This is one of the largest enterprise AI rollouts to date from a single company and signals that the era of "AI pilots" is ending — big companies are making full bets.
Why it matters: When a 30,000-person company standardizes on one AI agent platform, it creates lock-in, shapes workflows, and sends a loud signal to every enterprise still sitting on the fence.
Quick Hits
- →Canva shipped its first in-house image generation model, available to all users with promises of better customization and lower costs — a direct shot at Adobe Firefly's design-tool dominance. TechCrunch
- →OpenAI made its Jev-competing decision model widely available as the emerging "decision model" category gets crowded fast. TechCrunch
- →Meta says its Muse AI app now has 10M+ users, with shopping emerging as a top use case — Zuckerberg's agent-to-agent future is quietly taking shape. TechCrunch
- →Recursive launched Flint, a free laptop-sized coding model it claims was largely built by its own AI — an early, real-world test of recursive AI self-improvement. TechCrunch
- →Nirva raised $33M for a $249 AI pendant that only listens to you — because the wearable ambient AI race isn't slowing down. TechCrunch
AI TLDR