Anthropic has officially filed for its initial public offering, revealing staggering earnings alongside warnings of artificial intelligence risks. Meanwhile, a defense contractor recently plugged Anthropic’s Claude into a secure Department of Defense system in an unexpected security incident.
Anthropic Files for Historic IPO
Anthropic has publicly filed for its IPO, with expectations to raise more money than any previous company and become the first major US firm focused solely on AI software to go public. The company’s valuation reached $965 billion in its most recent round, up from $183 billion a year prior, and could approach $2 trillion once trading begins, as reported by WIRED’s Uncanny Valley podcast. The S-1 filing revealed extreme financial figures, including billions in revenue and losses, with revenue hitting $16.3 billion in the first half of 2026—a 14-fold increase from the previous year. Anthropic CEO Dario Amodei focuses heavily on enterprise coding products, which drive the company’s business growth.
However, the filing also highlighted vulnerabilities, such as nearly half of the revenue coming from just three tech companies that also act as competitors. Furthermore, the prospectus explicitly warned of existential risks, noting that advanced AI systems could pose catastrophic threats to humanity if not properly safeguarded. Interestingly, the same recursive self-improvement capability cited as a primary competitive advantage is also identified as a major existential risk factor.
Defense Contractor Connects AI Agent to Pentagon Network
In a separate report by WIRED’s Victoria Elliot, a major security incident occurred at the Department of Defense after a contractor connected a secure centralized system to a personal account for Claude Code, Anthropic’s commercial AI coding agent. The contractor, working for global consulting firm Accenture, used an agent that sent information about the DOD system and login credentials to a high-risk third-party website, operating inside the DOD network for two months undetected. The compromised centralized system contained controlled unclassified information, which, when combined with other data, can create classified information related to critical infrastructure, nuclear submarine propulsion, and biological or chemical terrorism vulnerabilities.
The Accenture employee automated the coding process by providing login credentials, allowing the agent to write and commit code directly to the system while connected to the internet. Anthropic’s Claude is currently barred from use by DOD staff. When contacted by WIRED, a DOD spokesperson stated that the department does not comment on network or system status, while an Accenture spokesperson maintained that the allegations were thoroughly investigated and found to be without merit, though they did not clarify who conducted the investigation.
As the midterm elections approach, grassroots organizations across the country are also preparing volunteer networks to document potential Immigration and Customs Enforcement deployments at polling sites and inform voters of their rights.
