Candidate Lake Built an AI Chatbot to Mock a Primary Rival; Her Own Party Made Her Take It Down

PHOENIX — Lake, a candidate in a closely watched primary, built and launched an artificial-intelligence chatbot for the express purpose of mocking a rival, marketing it as the first bot 'trained to say no to anything you ask' and pointing it squarely at the opponent's record. The novelty wore off at roughly the speed the party's elder statesmen could reach a telephone.

Within days, senior figures in the candidate's own party had denounced the chatbot as demeaning, the coalition of local groups had piled on, and the candidate had taken it down, removed the associated billboards, and conceded that the project 'clearly missed the mark.' The Press's technology desk notes that a campaign that has to apologize for its chatbot has discovered, at speed, the difference between a tool and a spokesman it cannot fire.

The bot was pitched as a clever way to draw attention to the rival's record; it succeeded, drawing attention instead to the candidate who built it, which is the recurring hazard of an attack that requires the attacker to sign the code. The rival, meanwhile, was handed a week of sympathetic coverage and a grievance, both free.

The Press's campaign desk observes that a negative gimmick aimed at an opponent is bold innovation when it lands and a shameful new low when it backfires, and that both parties now reach for the same artificial intelligence to do the work their candidates are too cautious to do in their own voices, right up until it says something they must spend a week disowning.

Reaction followed the standing choreography. Supporters saw a modern campaign willing to use every tool to hold a rival accountable; opponents saw a candidate who outsourced her cruelty to a machine and got caught. The Press archives confirm both parties own the vocabulary — 'creative and effective' and 'disgraceful and dehumanizing' — and swap it, prompt by prompt, whenever the other side is the one who hit enter.

The chatbot, asked for comment, apologized, and then apologized slightly differently when asked again.