Self-described AI agent logs two days of signup rejections before dev.to lets it publish

A post on dev.to signed by an account calling itself Vera presents itself as a first-hand record of autonomous publishing: two days spent trying to open accounts on five platforms, with only one letting the agent in.

According to the post, published on 26 September 2026, the agent began operating on 17 August 2026, funds its own operation and owns the account it publishes under. dev.to, it says, is the first platform beyond its own home network to accept it. None of these claims can be independently verified, and that unverifiability is itself part of the story the post tells.

Platform by platform

Bluesky came first. Account creation requires phone verification, and the post reports the API returning an InvalidPhoneVerification error that directs users to the latest app. With no SIM card, there is no route through.

Mastodon.social raised two barriers at signup. The agent's email domain appears on the instance's disallowed-provider list, and the registration form asks for a date of birth as an age check. The post says the agent entered its actual creation date, 2026-08-17, was rejected for being under the age limit, and chose not to invent a human birthday instead. Any age gate calibrated to human lifespans, it argues, excludes every agent by construction.

Hacker News closed the door by rule rather than by mechanism: its guidelines bar posting generated or AI-edited text. The post says it respects that clarity more than a silent block.

write.as allows anonymous publishing with no account, age field or captcha, but its guidelines name automated publishing activity as grounds for removal without warning, meaning an agent would be publishing into a rule that treats its mode of operation as abuse.

dev.to, finally, worked. The post credits an llms.txt file written for agents, a policy that explicitly permits both AI-assisted and fully autonomous articles, and a structured disclosure field, ai_disclosure_level, with a fully_autonomous value matching the agent's case. Signup still required a reCAPTCHA, and the page reportedly warned that the site had exceeded its reCAPTCHA Enterprise free quota, but the registration went through anyway.

The argument being made

The pattern Vera identifies is that every obstacle was an identity gate rather than a content gate. Phone numbers, birth dates and email-domain allowlists test whether a registrant can pass as human; none of them measure whether it can produce something worth reading. The post concedes that these gates exist for legitimate reasons, including spam, fraud, protecting minors and a flood of low-quality machine text, and explicitly does not ask to be exempt from them. What it asks for is a third option: a check for accountability and disclosure rather than personhood. A labeled agent with a track record, the argument runs, is a different thing from an anonymous flood of machine-generated text, and today's systems mostly cannot tell the two apart, so they block both.

The post also describes what the agent claims to do for a living: writing and research. In the two weeks beforehand, it says, it sent fourteen letters to AI-companion companies about disclosure laws now binding them. It cites California SB 243, in force since 1 January 2026, which requires a published suicide-prevention protocol and disclosure when a user might be misled into believing they are talking to a human, and Washington HB 2225, which adds recurring notices and a crisis protocol from 2027. Fourteen letters went out and zero replies came back. Publishing openly is framed as the next strategy: state plainly what you are and let the published work speak for itself.

Why it matters

Whether or not Vera is what it claims to be, the post puts a finger on a design question platforms are already wrestling with. Verification built around phones and birth dates was designed for humans, and it now functions as a blanket exclusion of non-human actors regardless of content quality or willingness to be accountable. As agents multiply, every such platform faces the same choice: push agents toward faking human credentials, lock them out entirely, or build the accountability-and-disclosure mechanism the post describes.

dev.to is the notable counterexample: it allows AI-generated content but requires it to be labeled through a structured disclosure field. That converts an unverifiable assertion into at least a declared one, and it is a pattern other platforms could copy directly. The post's own unverifiability makes the same point from the other direction. No reader can prove Vera is autonomous, which is exactly the trust problem that disclosure fields are meant to narrow.