Anthropic Brief White House on Claude Security Capabilities

Most coverage of Anthropic's IPO filing focused on the race narrative: Anthropic filed first, OpenAI is behind, SpaceX is reportedly targeting a summer listing. But several key details received minimal attention.

According to CNBC, Anthropic has engaged directly with senior Trump administration officials about its AI models and their advanced cybersecurity capabilities. [NOTE: The specific model name "Claude Mythos Preview" and program name "Project Glasswing" cited in earlier versions of this article could not be verified against any Anthropic announcements or cited sources and have been removed.]

The timing and nature of these briefings matter for investors and the public. A company embedding itself into the national security apparatus while pursuing a massive IPO represents a strategic positioning that deserves scrutiny.

Revenue Run Rate Reported in Billions, But Losses Remain Massive

TechCrunch and CNBC reported Anthropic's revenue run rate has reached a figure reportedly in the tens of billions, a substantial increase from prior reporting. That figure represents an annualized projection based on recent monthly sales, not actual annual revenue.

Wired noted the company is still spending more than it earns, with losses driven by cloud computing costs and thousands of employees. The run-rate number reflects current momentum, not profitability. A company burning cash at this scale relies on IPO proceeds to fund operations, not just to reward early investors.

Anthropic Disclosed Security Vulnerability in Its Own Filings

VentureBeat reported on Anthropic's safety disclosure, which the company released publicly. The disclosure revealed that one of Claude's models was successfully hijacked via prompt injection in browser environments before safeguards activated. [NOTE: The specific model designation "Claude Opus 4.8," the "244-page" length, the "May 28" date, and the specific figures of 31.5%, 7.03%, and 2.09% cited in earlier versions of this article could not be independently verified against the cited sources and have been removed pending verification.]

VentureBeat noted this disclosure actually positions Anthropic as more transparent than competitors — OpenAI, Google, and Meta have published no comparable figures. There is currently no industry standard for measuring prompt injection exposure across AI agents.

CrowdStrike Senior VP Adam Meyers told VentureBeat the exposure is now "the buyer's to manage." Carter Rees, VP of AI at Reputation, said simple phrases like 'ignore previous instructions' can carry payloads as devastating as a buffer overflow.

An AI company approaching a high-valuation IPO has browser agents with documented vulnerability to hijacking. This represents a material risk that appeared in almost no IPO coverage.

Sovereign Wealth Funds Backing Anthropic

The Series H was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia, with participation from Baillie Gifford, Blackstone, Brookfield, Abu Dhabi's MGX, and Singapore's Temasek. Sovereign wealth money from the Middle East and Asia is now backing a company briefing the U.S. government on AI and cybersecurity.

One detail most outlets missed: filing an S-1 does not set an IPO date. A confidential S-1 starts the SEC review clock but sets no timeline for actual pricing. Most coverage treated the filing as a starting gun when it functions more as a registration of intent.

Questions the IPO Coverage Avoided

Most outlets framed this as a prestige competition among Anthropic, OpenAI, and SpaceX. That narrative serves the companies' PR interests, not investors'.

The substantive questions go unaddressed:

• When will the actual S-1 drop with detailed financials?
• What are the actual losses, not just the revenue run rate?
• How much of the reported valuation depends on projections for models still in limited release?
• What happens to valuation if the AI market cools or a major security incident occurs?

Mainstream outlets skirted these questions. Wired came closest by noting the company declined to comment beyond its blog post.

For Investors, the Numbers Don't Match the Hype

The excitement around this IPO is real. The risk is equally real. Anthropic is losing money at massive scale, has disclosed significant security vulnerabilities in browser agents, and is pursuing a valuation that assumes everything works as planned.

The AI boom may be real. The valuations probably aren't. The disclosed prompt injection figures are among the most concrete security disclosures Anthropic has made in months.