Anthropic's Project Glasswing Finds More Than 10,000 Vulnerabilities

The first progress report on Project Glasswing, the cybersecurity initiative Anthropic announced last month, has been published. Working with roughly 50 partner companies, Anthropic says it used Claude Mythos Preview to uncover more than 10,000 high-severity and critical vulnerabilities in the world's most critical software.

Cloudflare alone detected 2,000 bugs (400 of them high or critical severity), and the Cloudflare team assessed that its false positive rate "outperforms human testers." Mozilla also found and fixed 271 vulnerabilities in Firefox 150 — more than ten times the number Claude Opus 4.6 discovered in Firefox 148.

The results in open source are particularly noteworthy. Anthropic scanned more than 1,000 open source projects and identified an estimated 6,202 high or critical vulnerabilities, and verification by independent security research organizations confirmed that 90.6% were true positives. As one example, a certificate forgery vulnerability in the cryptography library wolfSSL (CVE-2026-5194) was discovered and has already been patched.

That said, while "finding" vulnerabilities has become easier, the report also notes that the human resources for "fixing" them have become the bottleneck. On average, patching high or critical bugs is taking two weeks, and some maintainers are calling for the pace of disclosure to be adjusted.

The UK AI Security Institute reports that Mythos Preview is the "first model" to solve every step of a cyber exercise. XBOW likewise rates it as "far ahead of all existing models," signaling that AI's cybersecurity capabilities have entered a qualitatively new stage.

Anthropic Set to Close Funding Round of More Than $30 Billion

According to Bloomberg, Anthropic is expected to close a funding round of more than $30 billion as soon as next week. The valuation tops $900 billion, making it the world's most valuable AI startup ahead of OpenAI. Sequoia Capital, Dragoneer Investment Group, Altimeter Capital, and Greenoaks Capital Partners are co-leading the round, with each planning to invest around $2 billion. Existing investors including Founders Fund and General Catalyst are also participating.

Cloudflare Cuts More Than 20% of Its Workforce, Citing AI

Cloudflare CEO Matthew Prince has laid off more than 20% of the company's employees despite record revenue. Prince claims that "AI will replace middle management and compliance departments," but according to The Decoder's analysis, the company has presented no concrete evidence — and some see the move as a conventional efficiency drive wrapped in AI branding, set against declining profit margins and 40% headcount growth over two years.

Lawyers Keep Citing AI-Generated "Fake Case Law"

Scientific American reports that there is no end to cases in which AI generates fabricated court decisions and lawyers then cite them in legal documents. The real-world harm of AI hallucination in the legal industry is becoming serious, and establishing guidelines for the use of AI tools has become an urgent priority.

NTSB Withdraws Investigation Materials Over AI "Voice Recreation of the Dead"

According to Ars Technica, the U.S. National Transportation Safety Board (NTSB) withdrew the relevant investigation documents after a case arose in which AI was used to recreate the voice of a pilot who died in a crash, based on the crash investigation materials. The incident is once again calling the ethical boundaries of AI voice cloning technology into question.

AI Startups' "Inflated ARR" Becomes a Problem

TechCrunch has reported on the reality that some AI startups are arbitrarily stretching the interpretation of traditional revenue metrics, and that investors are knowingly encouraging the practice. Concerns are mounting over the soundness of the AI bubble.

---