Anthropic's opaque 'suspicious signals' bans hit paying Claude users

A first-person account of losing Claude access over unspecified "suspicious signals" drew front-page attention on Hacker News, with its author describing how a paid Anthropic account was revoked without warning, cause or a way to reach a human.

According to the post, published on 2 September at kix.codes, Anthropic's Safeguards Team sent a templated email stating that an internal investigation had found "suspicious signals" indicating a Usage Policy violation. No clause was named and no behaviour was described. The only appeal route, the author writes, is an in-product form that appears after logging back in — there is no one to make a case to.

The stakes were significant: he says he had paid for Claude for years and was on the $200-per-month Max plan, using the account as his primary daily tool. He adds that a colleague in the Philippines on the $100 Max tier was banned immediately after paying, and that other reports he has seen describe bans landing close to billing events. He explicitly avoids claiming a cause — only that having just paid is clearly no protection.

A cluster of similar reports

The post points to Claude Code's public issue tracker, where several complaints follow the same shape. A marketer in Brazil hit a "banned organization" error mid-checkout and then received the same Safeguards email despite, in their telling, ordinary coding and content work (issue #40046). A long-time customer operating across Singapore and Malaysia upgraded to Max and Team plans and saw every related account suspended roughly eleven hours later with the same notice (#45936). Another $200 Max user on the official CLI reported the identical template, then around two months of silence on appeal (#51670).

Outcomes vary. The author cites a founder whose account was flagged after travel to India and restored within days. But fast reversals, he argues, are the exception; the default experience is an automated notice, an opaque process and an open-ended wait.

Enforcement without explanation

The author concedes Anthropic has real abuse to police, pointing to an enforcement wave earlier in the year against third-party harnesses — OpenCode among them — that used consumer OAuth credentials as an unofficial API. His objection is narrower. Anthropic's Help Center, he notes, lists enforcement categories such as repeated Usage Policy violations, unsupported locations and terms-of-service breaches, yet nothing defines what a "suspicious signal" is, and the revocation email does not either. A legitimate user cannot tell whether a location heuristic, a payment check or something else tripped the system, and so cannot correct it.

He also contrasts leverage. Companies spending heavily through the API have invoices, account managers and commercial weight; an individual subscriber, even after years of payments, gets a no-reply address and a reference number, and loses their working tools for as long as the appeal takes. At publication time the appeal was filed and unresolved, and the post reports no response from Anthropic beyond the initial notice.

Why it matters

One unverified blog post would matter little on its own, but this one matches a documented pattern in public issue trackers and describes a severe failure mode: sudden, total loss of a paid, daily-use product with no stated cause and no human contact. For anyone running their work on a single frontier lab's consumer offering, it reads as a case study in single-vendor risk — the author says he now intends to spread his workflow across providers, open-weight models included, rather than depend on one company's automated enforcement. It also tests Anthropic's safety-first positioning: opaque, unaccountable account enforcement undercuts the careful-operator brand. Per the post, Anthropic has yet to explain what these signals are, how they are detected, or how a paying customer can avoid setting them off.