DOJ Appeals Pentagon Ban, Claude Code Leak, API Crackdown

Three separate developments landed for Anthropic this week, each touching a different part of the business: a federal legal fight re-escalates, an internal document surfaces publicly by accident, and the company tightens control over how subscribers access Claude.

Key points

• The Justice Department filed an appeal to reverse a March 26 injunction that had blocked the Trump administration’s ban on government use of Anthropic products.
• Anthropic accidentally exposed the system-prompt instructions behind its Claude Code AI agent, offering a rare public look at the model’s internal construction.
• The company cut off support for OpenClaw, a third-party client, citing concerns about how such tools consume Claude subscription access.

What does the DOJ appeal mean for the Pentagon injunction?

The March 26 ruling was a win for Anthropic: a federal judge blocked the Trump administration from banning government agencies from using its AI. The DOJ’s decision to appeal, reported by PYMNTS on April 2, signals the administration is not prepared to accept that outcome and will push to reinstate the prohibition at the appellate level.

For Anthropic, the government sales channel remains in legal limbo. Revenue from federal contracts, a meaningful part of any enterprise AI growth story, cannot be fully counted until the circuit court rules. The company won the first round, but appellate proceedings could take months. Investors watching the IPO timeline should note that a Q4 listing, which sources flagged last week, would likely occur before this legal track resolves.

The administration’s national security arguments were rejected at the district level. Whether an appellate panel views the same record differently is genuinely uncertain. Both sides have made substantive claims, and the outcome will shape not just Anthropic’s access to federal business but set a precedent for how the executive branch can restrict specific AI vendors from government use.

How significant is the Claude Code prompt leak?

Anthropic inadvertently published the instructions that govern Claude Code, the company’s AI coding agent, according to Benzinga’s April 1 report. The disclosure appears to have been unintentional. System prompts are among the more closely guarded parts of any commercial AI product because they encode the behavioral rules and persona that differentiate one vendor’s model from another.

The practical competitive damage depends on how specific and transferable those instructions are. Competitors can read them, but replicating the underlying model weights is a separate and much harder problem. Still, the leak removes a layer of opacity that Anthropic, like every major AI lab, generally prefers to maintain. It also raises operational security questions that institutional customers and prospective IPO investors will reasonably flag.

No public statement from Anthropic on the cause or scope of the disclosure was included in the source material reviewed for this edition.

Why is Anthropic cutting off third-party clients?

Business Insider reported on April 4 that Anthropic is ending subscription-level support for OpenClaw, a third-party tool that routes Claude access through the standard consumer subscription rather than the API. Anthropic’s stated concern is that such tools place disproportionate load on subscription infrastructure and, by extension, undercut the economics of its tiered pricing model.

This move is consistent with a broader pattern among AI companies: as usage scales, the gap between what a flat subscription permits and what a power user extracts through third-party clients becomes a real cost problem. Forcing higher-volume use cases toward the API, where consumption is metered, is straightforward margin management.

For retail Claude subscribers, the immediate effect is that OpenClaw stops working as a front-end for their accounts. For Anthropic, the change brings API revenue recognition more in line with actual compute costs. Neither outcome is dramatic, but together the three stories this week sketch a company managing growth friction across legal, operational, and commercial dimensions simultaneously, all while an IPO clock ticks in the background.