Detailed Analysis
A developer posting to Hacker News describes being permanently banned from Anthropic's Claude Code service twice within approximately one month, despite using the tool for routine software development tasks such as summarizing documentation and refactoring code modules. The user reports that their first ban occurred after roughly one hour of usage, which they attribute to having an active VPN connection at the time of signup or use. After receiving a $120 refund, they attempted to create a new account without a VPN, only to be banned again — this time, they believe, because they used the same credit card linked to the previously banned account. Anthropic's support response consisted of a generic policy-violation notice with no specific explanation or path to appeal, prompting the user to turn to Hacker News in hopes that an Anthropic employee would intervene directly.
The case illustrates a structural tension that has become increasingly visible as AI developer tools scale rapidly: automated fraud and abuse detection systems that are calibrated for risk prevention at scale can produce significant false positives, disproportionately affecting legitimate users who happen to trigger behavioral or account-linkage signals. VPN usage and shared payment instruments are well-established proxies for suspicious activity in consumer platforms, but they are also extremely common among developers who routinely use privacy tools for professional or security reasons. When these heuristics produce bans without transparent explanation or meaningful human review, the result is a customer experience failure that can cause real harm to individuals relying on these tools for professional work.
The broader context matters here. Claude Code, Anthropic's terminal-based agentic coding assistant, has been positioned as a premium professional product competing directly with GitHub Copilot, Cursor, and other AI coding environments. The product's value proposition depends heavily on trust and continuity — developers invest time integrating these tools into their workflows, and abrupt, unexplained account terminations undermine that proposition acutely. That the user had been accessing Claude Code through their employer for months before attempting personal use adds an additional dimension: the same individual was presumably a trusted enterprise user in one context and a flagged bad actor in another, solely due to account-level signals rather than behavioral ones.
The incident also reflects a broader industry-wide challenge in AI service governance. As companies like Anthropic, OpenAI, and Google scale their API and consumer AI products, they face mounting pressure to enforce usage policies against misuse — including account farming, prompt injection at scale, and attempts to extract harmful outputs — while simultaneously maintaining accessible, low-friction experiences for genuine users. The absence of a transparent, human-reviewable appeals process, as evidenced by the generic support response in this case, represents a governance gap that becomes more consequential as these tools move from experimental novelties to critical professional infrastructure. The user's recourse to a public forum as a last resort underscores how inadequate current accountability mechanisms remain for individuals caught in automated enforcement systems.
Read original article →