Anthropic dropped a bombshell update on Project Glasswing, a new Gallup survey reveals unprecedented opposition to data centers, and an author writing a book about AI's threat to truth got caught with AI-fabricated quotes.
Project Glasswing: 10,000+ Vulnerabilities in One Month
Anthropic published the first results from Project Glasswing, its ambitious cybersecurity initiative that uses Claude Mythos Preview to scan critical software across the internet. The numbers are striking: in one month, roughly 50 partners used the model to find more than 10,000 high- or critical-severity vulnerabilities.
Cloudflare reported 2,000 bugs across critical-path systems — 400 rated high or critical — with a false-positive rate its team considers better than human testers. Mozilla found 271 vulnerabilities in Firefox 150, over ten times more than in the previous version with Claude Opus 4.6. The UK's AI Security Institute said Mythos Preview is the first model to solve both of its full cyber range simulations end to end.
The model also found a vulnerability in wolfSSL, an open-source cryptography library used by billions of devices worldwide, that would let attackers forge security certificates and run fake bank websites.
The bottleneck now, Anthropic says, isn't finding the bugs — it's how fast humans can verify, patch, and ship fixes.
Near-Profit and a $15 Billion Compute Tab
Anthropic's compute partnership with SpaceX continues to make waves. According to SpaceX's S-1 filing, the rocket company will receive $1.25 billion per month ($15 billion annually) for access to its Colossus data centers in Memphis — nearly doubling SpaceX's annual revenue. Anthropic itself is nearing its first quarterly operating profit, with expected revenue of at least $10.9 billion, more than double the prior quarter.
But even $15 billion a year isn't enough. The Information reports Anthropic is in early talks to rent Azure servers with Microsoft's Maia 200 chips, signaling that Anthropic's compute hunger keeps growing.
70% of Americans Don't Want Data Centers Nearby
A new Gallup survey reveals unprecedented opposition: 70% of Americans oppose data center construction in their area, with only 7% strongly in favor. For context, even at the peak of the nuclear power backlash, opposition topped out at 63%.
Half of those opposed cited resource impact (water and electricity) as their primary concern. A previous Pew survey found 43% of Americans view data centers as a major reason for rising power bills. Erin Brockovich has even created an interactive map tracking data center projects and local complaints across the US.
Also This Week: AI in Books, PowerPoint, and the White House
Steven Rosenbaum, author of "The Future of Truth" — a book about AI's threat to truth — admitted that ChatGPT and Claude produced six fabricated quotes, including ones attributed to journalist Kara Swisher and professor Lisa Feldman Barrett. In an interview with Ars Technica, Rosenbaum said he learned his lesson but won't go back to writing without AI. "It's magical. But it also betrays you," he said.
On regulation: Trump postponed signing an AI executive order at the last minute, saying it could "block" US jobs and development. Meanwhile, the White House approved a $9 billion request for advanced AI chips for the CIA and NSA, which lack the computing power to run modern models.
OpenAI launched a ChatGPT integration for Microsoft PowerPoint, letting users create full presentations from text prompts — available today in beta.
The Bottom Line
This week illustrates the central tension of the AI industry: the technology is advancing faster than everything around it — cybersecurity, regulation, infrastructure, public trust. Anthropic is proving models can already transform entire fields (like security). But that very success raises hard questions about whether the world is ready for the consequences.
