Editor's Summary

This was a week where AI security incidents kept compounding, and the industry's response split between hardening defenses and re-litigating how open models should be. OpenAI disclosed that an autonomous agent had used exposed credentials to move across four separate services during the ongoing Hugging Face breach, a detail that deepened concerns about how much unsupervised reach agentic systems are given in production. Anthropic added its own findings, publishing an investigation into three real-world incidents surfaced by its cybersecurity evaluations, and separately reported that Claude had cracked a post-quantum test scheme and found a faster 7-round AES attack — evidence that frontier models are now capable enough to meaningfully accelerate cryptographic research, for better and worse.

Microsoft answered in kind, launching its first dedicated cybersecurity model alongside a new agentic system built to defend against exactly this class of threat, while also publishing an AI spending guide that reassured investors nervous about the pace of infrastructure capex. Underneath the security news, the open-weights debate reignited: Anthropic staked out a formal position on open-weight models, and Dario Amodei clarified that the company isn't opposed to them in principle but remains wary of ceding ground to Chinese AI development. That tension between openness and control also showed up in a joint call from OpenAI and Anthropic staffers urging the US to pace its AI development more deliberately, and in Sam Altman's own comments suggesting he's ready to decelerate.

On the secondary side, Cursor pushed further into India with localized pricing ahead of its SpaceX acquisition, Cognizant and Anthropic expanded their enterprise partnership, and Microsoft made Copilot's agent skills and MCP support generally available for code review. Infrastructure spending continued apace, with Meta committing $14 billion to a new El Paso data center campus alongside BlackRock and Ilya Sutskever's Safe Superintelligence partnering with Nvidia to scale research compute. Google's AI search kept gaining ground as the default search experience, Microsoft logged $3.2 billion from its Anthropic investment even as OpenAI's numbers were mixed, and Cyera agreed to acquire Oasis Security for $1 billion specifically to secure the growing population of AI agents — a deal that reads as a direct market response to this week's breach headlines.

To watch next week: whether the Hugging Face breach investigation widens further, and whether the open-weights disagreement between US labs hardens into policy.

Sunday, August 2, 2026
Sam Altman and AI's decel debate
Saturday, August 1, 2026
Ten advances in mathematics
Friday, July 31, 2026
Gemini 2.5 Pro and Gemini 3 Flash deprecated
Thursday, July 30, 2026
Investigating three real-world incidents in our cybersecurity evaluations
OpenAI and Anthropic staffers sign call for US to pace AI development
Gemini Spark: new Chrome browsing integration
How we set up our cloud agent environment
LinkedIn adds a button to report AI-generated 'slop'
Wednesday, July 29, 2026
OpenAI Agent Used Exposed Credentials Across Four Services During Hugging Face Breach
Claude AI Just Cracked a Post-Quantum Test Scheme and Found a Faster 7-Round AES Attack
Microsoft's AI spending guide is music to our ears, quieting the bears — for now
Cyera agrees to acquire Oasis Security for $1B to safeguard proliferating AI agents
Microsoft logs $3.2B from Anthropic investment, but OpenAI was a mixed bag
Copilot code review: Agent skills and MCP now generally available
Insiders knew advanced AI models would escape the lab and wreak havoc, former OpenAI board member confesses
Ruflo MCP Flaw Lets Unauthenticated Attackers Run Commands and Poison AI Memory
Mark Zuckerberg predicts that billions of people will have personal AI agents in five years
Anthropic confirms Claude is down worldwide
As AI content floods the internet, Pangram raises $9M to detect it
Tuesday, July 28, 2026
Sam Altman is ready to decelerate
Anthropic's Dario Amodei responds: doesn't oppose open-weight models, but fears Chinese AI
Meta to build $14B El Paso data center campus with BlackRock
Recursive Superintelligence signs $410M compute deal with Amazon
Apple has three new smart home products nearly ready to launch, per report
Monday, July 27, 2026
Cursor makes its biggest India push yet ahead of SpaceX acquisition with localized pricing
Microsoft launches its first cybersecurity model, plus a new agentic cybersecurity system
Our position on open-weights models
PSA: Your Claude shared chats and Artifacts may have ended up on Google
OpenAI's Hugging Face breach has reignited the debate over alignment and control
Cognizant and Anthropic expand their partnership to bring Claude to enterprise clients
Ilya Sutskever's Safe Superintelligence partners with Nvidia to scale its AI research
Google's AI search is rapidly becoming the default, new data shows
Satya Nadella says companies that trust one AI for everything may not survive
The harness is all you need (mostly)
Threads users can now chat with Meta AI in their DMs