Anthropic 代码泄露事件分析

Anthropic accidentally released the internal source code of Claude Code due to human operational error, involving approximately 1,900 files and 512,000 lines of code. It subsequently used a GitHub copyright takedown to remove copies on a large scale, but mistakenly affected more repositories in the process. The company is modifying its automated processes to prevent recurrence. A few days earlier, media also reported that the company had exposed thousands of internal documents on a publicly accessible system, including draft information about the upcoming model "Mythos/Capybara".

This incident occurred against two sensitive backgrounds:

• Anthropic is currently in a legal dispute with the U.S. government over the Pentagon's decision to label it as a supply-chain risk. The company claims this label could lead to billions of dollars in revenue loss.
• Claude Code is a key product for the company's recent growth and commercialization. The company disclosed that its run-rate revenue as of February had exceeded $2.5 billion, supporting its goal of an IPO as early as this year.

Source Quality:

• Primary Information: Anthropic official statement, remarks by Claude Code lead Boris Cherny on X, GitHub takedown notice.
• Secondary Information: Bloomberg summary report, Fortune's report on the previous document exposure incident.
• Credibility: Medium to High. The incident itself is credible, but quantification of revenue loss and competitive consequences remains largely speculative.

【Most Valuable Incremental Information】

• Claude Code's run-rate revenue exceeded $2.5 billion as of February. This is the most crucial incremental fact, indicating the leak hit not a peripheral project but Anthropic's most valuable current commercial asset.
• This is the "second information exposure within days," escalating the issue from a single incident to questions about process and governance capabilities.
• The company is currently entangled in a dispute with the Pentagon over its supply-chain risk designation. Security and governance incidents will significantly weaken its persuasiveness with government and high-compliance industries.
• The leaked content may allow outsiders to see unreleased features, such as the resident agent "Kairos" and a user frustration tracking system, leading to premature product roadmap exposure and public opinion risk.
• Competitors no longer need high-cost reverse engineering to understand Claude Code's internal implementation and protection strategies faster, lowering the barrier to competition and attack.

【Expectation Gap Judgment】
The market previously more easily interpreted Anthropic's core narrative as: strong model capabilities, Claude Code commercialization exceeding expectations, nearing an IPO, and being one of the strongest AI application monetizers besides OpenAI. The expectation gap brought by this news is that the market may have underestimated the damage of "governance shortcomings" to valuation and government business.

More specifically:

• For consumer users, the leak may not immediately cause churn, as no customer data was exposed.
• For enterprise and government clients, consecutive internal exposures will significantly increase audit, procurement, and legal resistance.
• For IPO investors, the event will shift focus from high growth to "whether internal controls are mature" and "whether it can withstand public company-level scrutiny."

Therefore, this is not a typical short-term product incident but more of a governance discount event.

【Impact Path】

Revenue
In the short term, individual developer subscriptions and general API usage may not be significantly affected; however, large enterprise, financial, healthcare, government, and other high-compliance clients may lengthen procurement cycles, slowing Claude Code's expansion.

Profit Margin & Cash Flow
The company needs to invest more in security, audit, release management, and compliance resources. Sales cycles may also lengthen, meaning customer acquisition and support costs will rise, indirectly compressing operating leverage.

Valuation
If there are no further material losses, the valuation impact will mainly manifest as a multiple discount rather than a significant downward revision of revenue forecasts. If the Pentagon dispute worsens or customer churn occurs, it will shift to a fundamental downgrade.

Competitive Landscape
Competitors that place greater emphasis on enterprise-level governance and development process security will benefit relatively. In the public market, MSFT can more easily strengthen the narrative of "enterprise-grade security and process maturity" through GitHub Copilot. In the private market, other AI coding agent vendors will also seize the opportunity to compete for high-end clients.

【Trading Implications】
Direct tradability is limited as Anthropic is not publicly listed. More realistic mapping approaches fall into three categories:

• AMZN, GOOGL: As important partners/investors in Anthropic, there may be slight sentiment disturbance in the short term, but given the scale of these two companies, it is not enough to constitute a strong trading signal on its own.
• MSFT: The relative beneficiary logic is clearer. If enterprise clients develop doubts about Anthropic's governance capabilities, the credibility of GitHub Copilot and Microsoft's enterprise security system will become a stronger selling point.
• Cybersecurity & Compliance Sector: Medium-to-long term beneficiary of increased investment by AI software companies in code release, permission management, audit, and data protection. However, this is more like a sector beta, not an alpha driven by a single news piece.

In conclusion, this piece is more suitable for making "relative strength" and "valuation discount" judgments, not for serving as a directional heavy position basis for parent company holdings.

【Risks & Uncertainties】

• There is currently no evidence of customer data or credential leakage, so direct business damage may be lower than headline intuition suggests.
• Whether the code being studied by outsiders will lead to exploits, bypass protections, or competitive replication still requires time to observe.
• The $2.5 billion run-rate is the company's statement, not equivalent to confirmed revenue, nor does it guarantee sustained growth.
• If Anthropic quickly fixes its processes, experiences no customer churn, and the government dispute does not worsen, the event's impact may be limited to short-term public opinion.

【Follow-up Questions to Verify】

• Whether clear signals emerge of clients pausing purchases, delaying renewals, or facing restricted eligibility for government contracts.
• Whether the Pentagon supply-chain risk dispute worsens further due to this incident.
• Whether Claude Code's active users, paid conversion, or enterprise seat expansion slow in the next 1-2 quarters.
• Whether the leaked code truly contains exploitable security weaknesses, not just product roadmap exposure.
• Whether Anthropic delays its IPO timeline or strengthens risk disclosures in its prospectus materials.