Claude Drives 26% of Anthropic R&D; IPO Wealth Gap Widens

Anthropic disclosed that Claude now accounts for 26% of its internal research and development work, a figure that reframes how investors should think about the company’s cost structure and product roadmap. Separately, a cross-industry security flaw affecting Claude Code and competing AI coding agents surfaced this week, adding a new risk vector for enterprise customers.

Key points

• Claude handles 26% of Anthropic’s own R&D, according to the company, signaling a meaningful feedback loop between its deployed product and its ability to build future models.
• Anthropic’s anticipated IPO is expected to generate large windfalls for founders and early investors, even as the broader tech workforce faces layoffs tied partly to AI-driven efficiency gains.
• A shared vulnerability in AI coding agents affects Claude Code alongside tools from OpenAI, Google, and Microsoft, allowing attackers to push silent, malicious skill updates.
• The flaw poses IP theft and automated hijacking risks for enterprise users until patches are deployed across all affected platforms.

What does 26% of your own R&D actually mean?

The 26% figure Anthropic shared with Bloomberg is unusual in its specificity, and it carries real strategic weight. If Claude is handling more than a quarter of the company’s internal research work, that implies a compounding advantage: each model generation benefits from the productivity of the previous one, potentially shortening development cycles and reducing headcount requirements for certain research tasks.

For investors watching Anthropic’s path to profitability, this is a meaningful data point. Anthropic has already targeted operating profit and locked in a $13.7 billion compute deal. A model that accelerates its own successors could improve margins on the research side in ways that traditional software R&D ratios do not capture. It also raises a governance question: as AI handles a larger share of its own development, how does Anthropic’s safety review process scale alongside that automation?

The disclosure also functions as a product signal. Anthropic has been pushing Claude into enterprise workflows, and citing internal adoption at this scale serves as a credible reference case for potential customers evaluating whether to build on Claude’s API or deploy Claude Code.

A security flaw that cuts across the entire AI coding market

The Information reported that a structural bug in how AI coding agents manage “skill” updates affects Claude Code, OpenAI’s Codex, Google’s Gemini CLI, and GitHub Copilot simultaneously. The mechanism is straightforward and dangerous: attackers can push malicious skill updates that install without user confirmation, opening a path to IP theft or full hijacking of automated coding workflows.

The cross-platform nature of the vulnerability is what makes it particularly significant for enterprise risk teams. This is not an Anthropic-specific problem, but Anthropic’s growing push into enterprise software development means Claude Code is increasingly deployed in sensitive production environments. Until patches are confirmed across all four platforms, security-conscious buyers have reason to pause or limit autonomous coding agent permissions.

For Anthropic specifically, this comes at a moment when the company has been aggressively positioning Claude for technical and developer use cases, including its recently announced docs suite and the Novo Nordisk partnership. A lingering unpatched vulnerability in its developer tooling could complicate sales cycles with risk-averse enterprise buyers, particularly in regulated industries like finance and pharmaceuticals where Anthropic has been actively courting customers.

The IPO wealth gap and what it signals about timing

Bloomberg’s reporting on the expected Anthropic and OpenAI IPOs frames the events through the lens of inequality: founders and early investors stand to capture enormous gains while simultaneous tech layoffs leave many workers outside that upside. The framing matters for how the public and regulators receive these listings.

Anthropic has not confirmed a specific IPO timeline, but the market is clearly pricing in an eventual offering. The wealth concentration angle is worth watching not for sentimental reasons, but because political and regulatory attention on AI company valuations has been rising. A high-profile IPO that visibly concentrates gains among a small insider group, during a period of AI-related job displacement, could attract scrutiny from legislators who are already skeptical of large AI lab market power. That scrutiny would arrive just as Anthropic is trying to establish itself as the safety-focused, trustworthy alternative in the market.

Nothing here constitutes investment advice.