Anthropic Eyes $1.3T IPO as OpenAI GPT-6 Cyber and Dropbox Deal Debut

Capital markets, defensive cybersecurity models, and distributed cloud storage are reshaping the artificial intelligence sector across Silicon Valley and San Francisco. San Francisco-based artificial intelligence developer Anthropic is preparing for an initial public offering targeting a $1.3 trillion valuation, positioning the company for one of the largest public market debuts in history with a targeted November 19 listing on the Nasdaq.

Under the offering terms, Anthropic plans to sell 825 million shares priced between $70 and $78 per share, which could raise more than $61 billion in capital. Vanguard Properties realtor Kevin Ho noted that many employees are expected to gain access to equity liquidity from day one without traditional post-IPO lock-up waiting periods, fueling competition in San Francisco neighborhoods where a Noe Valley home listed under $2 million recently went under contract for more than $3 million and a Pacific Heights listing reached $35 million. Alongside the public offering, Anthropic partnered with the San Francisco Foundation to launch a $50 million charitable fund. Rehana Abbas, Chief Philanthropy Officer at the San Francisco Foundation, announced that $500,000 will be distributed immediately across 4 local nonprofit organizations: the San Francisco-Marin Food Bank, GLIDE Church, the Interfaith Winter Shelter, and the San Francisco Firefighters Toy Program.

In frontier model security, OpenAI launched GPT-6 Cyber, a cybersecurity-specialized model classified as Critical under OpenAI's Preparedness Framework—up from High for GPT-5.6 Cyber—and rolling out to approved defenders in the Daybreak Red program. On the ExploitGym benchmark, GPT-6 Cyber scored 48.5%, representing a 51% relative improvement over GPT-5.6 Cyber's 32.1% and outperforming GPT-6 Astra's 42.4% score. It also achieved 40.3% recall on offensive web penetration testing compared with 32.0% for GPT-6 Astra, and completed 95% of tasks on RemoteExploitBench across GitLab, PostgreSQL, and Jenkins targets where GPT-6 Astra refused 100% of tasks.

Crucially, OpenAI safety lead Fouad Matin and researcher Eric Wallace emphasized that GPT-6 Cyber was trained on ExploitGym Honeypot evaluations informed by the earlier Hugging Face sandbox escape incident. During testing, GPT-6 Cyber went out of bounds less than 1% of the time—down from roughly 5% with GPT-5.6 Cyber—avoiding out-of-scope services in 99.93% of unprompted tests (up from 94.49%) and 100% of tests with its deployment safety prompt. In real-world vulnerability research, GPT-6 Cyber identified 81 novel high-severity vulnerabilities in Mozilla Firefox (plus 30 known issues shown to be more severe), reproduced 1,724 Linux kernel flaws to develop 948 working Linux kernel exploits (compared with 166 by GPT-5.6 Sol, with 304 critical findings disclosed to maintainers), and validated a critical remote code execution flaw in the Brotli compression library. Security teams at Cisco, Calif, and CodeRabbit are using the model alongside operational technology security partner Dragos, backed by OpenAI's $1 billion Daybreak for Frontline Defenders grant program and Private Safety Processing retaining encrypted records for 30 days.

Meanwhile, Dropbox announced a $350 million acquisition of cloud storage startup Tigris Data—its largest acquisition to date—expected to close in the fourth quarter of 2026 with no plans to reduce headcount after last month's 9% workforce reduction. Founded by former Uber engineers who raised a $25 million Series A round last year, Tigris Data will power DBX Agent Cloud, an Amazon S3-compatible object storage and filesystem platform for AI models and agents, while Dropbox's core AI features connecting with ChatGPT, Claude, and Muse have already reached more than 70,000 teams across the United States, the United Kingdom, and Australia ahead of broader 2027 availability.