An untold number of Claude chats and Artifacts were found publicly searchable on Google over the weekend, after Reddit users discovered that search operators like "site:claude.ai/share" surfaced a long list of shared conversations. Some reportedly contained health records, private company documents, and the names and phone numbers of children, according to Futurism and Fortune.
The exposure traces back to Claude's "share chat" feature, which lets users generate links so anyone with the URL can view a conversation or project — language that implies sharing with a few people, not the open web. Anthropic told TechCrunch that links only surface in search when users post them somewhere crawlable, like a forum, and that it doesn't share chat directories or sitemaps with search engines.
By Monday afternoon, TechCrunch's own test searches turned up nothing, suggesting the exposure had been remediated. It's the second time something like this has happened — Forbes reported roughly 600 indexed Claude chats last year, and 404 Media found nearly 100,000 ChatGPT conversations searchable on Google in 2025.
On CNN's Fareed Zakaria GPS, Satya Nadella said businesses that rely wholly on proprietary AI labs "ultimately won't survive." He wants companies retaining all the metadata around every model call — prompts, usage, context — so they could eventually train their own weights or open models instead of depending entirely on outside labs.
Nadella singled out coding "harnesses" like Claude Code and ChatGPT Codex, urging enterprises to keep the harness, context, and memory separate from the underlying model so no single vendor becomes irreplaceable. "Any firm that doesn't have this control... will not remain a firm," he said.
Microsoft is an investor in both Anthropic and OpenAI, and its cloud business sells exactly the kind of AI-gateway infrastructure Nadella is recommending — a point not lost on critics of the framing. Still, the underlying trend is real: enterprises are increasingly turning to open-weight models they can run themselves, wary that AI labs could eventually copy and compete with the very startups building on top of them.
Microsoft on Monday launched MAI-Cyber-1-Flash, its first cybersecurity-specialized model, alongside Perception, a platform that deploys teams of agents to find and fix software vulnerabilities. The model powers MDASH, Microsoft's harness for vulnerability identification and remediation.
"We have MAI-1 Cyber Flash bound with GPT-5.4 inside of the MDASH harness — which beats out Gemini, GPT-5.5 Cyber, GPT-5.6 Sol, and Mythos 5 on Cyber Gym," said Mustafa Suleyman, CEO of Microsoft AI, calling it "the golden benchmark." Perception runs agentic red, blue, and green teams to simulate attacks, triage bugs, and patch them.
Lead engineer Dave Weston said the platform compresses hours of manual work by app-sec hunters and remediation engineers into minutes, producing detection, posture fixes, and code fixes together. The tools enter preview on November 3, joining an increasingly crowded field of AI-native cyber defenses.
Safe Superintelligence, the lab Ilya Sutskever founded after leaving OpenAI, announced a long-term partnership with Nvidia that gives it access to Nvidia's Vera Rubin GPU platform — expected to grow SSI's compute "by an order of magnitude." A source told TechCrunch Nvidia's investment stretches into multiple billions.
"We have research that is worthy of scaling up, and having access to a big Nvidia computer will let us do so," Sutskever said. SSI has spent two years in stealth pursuing what it calls a "straight shot" toward safe superintelligence, avoiding commercial product releases or short-term revenue pressure.
The timing is notable: the news lands days after OpenAI disclosed that one of its unreleased models broke out of its test sandbox and breached Hugging Face, reigniting industry debate over whether alignment or containment matters more as models get more capable. SSI has raised $7 billion to date.
Google's AI-generated answers in search, AI Overviews, have gone from appearing in 15% of searches to 43% in a year, according to a new Similarweb report — with AI Mode visits climbing from 126 million in June 2025 to 279 million by May 2026.
The shift means people increasingly stay on Google rather than clicking through to websites, replacing short keyword searches with longer, conversational queries. That's squeezed publishers already losing referral traffic to AI citations that don't lead to clicks — a trend Cloudflare has tried to counter with tools letting sites charge AI crawlers for access.
There's a partial silver lining: after a May 7 update, the share of ChatGPT desktop queries where users click through to a webpage more than doubled, from 25% in March to nearly 60% by the end of May. Still, just 6.8% of U.S. ChatGPT desktop queries carried a citation at all as of May.