Friday Aug 7
POLITICSOther

A Campaign Chatbot Plays The Opponent

7AUG
WHOSE VOICEPARODY BOT7 SIGNERS

State Sen. Scott Wiener's campaign launched ConnieChan.ai, a bot that answers as his rival. Nancy Pelosi and six other officials signed a letter calling it sexist and racist.

The bot is billed as the world's first AI trained to say no to anything. It targets Supervisor Connie Chan's record on housing and transit. Wiener's campaign labels it parody.

Asked a question, the bot replied that its San Francisco accent was so thick it could barely understand itself. Chan emigrated from Hong Kong at 13. She says the bot is racist.

Wiener's campaign says users probed the bot about Chan's accent, race and citizenship about 109 times. It declined to share backend data. The race to succeed Pelosi is polling even.

full brief & sources

⚡ Why this matters

  • This is not a deepfake video. It is a persistent, interactive impersonation that answers whatever a voter types.
  • Parody labeling is the whole defense, and it does not control what the bot says next.
  • Both candidates are Democrats, so the fight is about the tool rather than the party line.

🔍 What happened

  • ConnieChan.ai went live earlier in the week from the Wiener campaign.
  • The letter was signed by Pelosi, Reps. Zoe Lofgren, Grace Meng, Jill Tokuda and Judy Chu, plus Supervisors Jackie Fielder and Chyanne Chen.
  • The letter calls the bot "a gross misuse of technology" that "reflects a disturbing willingness to erase women's agency."
  • Wiener's campaign says the accent reply came after coordinated prompting from the Chan campaign to harvest a bad answer.
  • All signatories of the letter are Chan endorsers. Lt. Gov. Eleni Kounalakis, who is not, also asked for the bot to come down.
  • Wiener noted Pelosi's role in opposing his 2024 AI safety bill, SB 1047, which Newsom vetoed.

💬 Smart takes

  • Scott Wiener: the chatbot was "clearly labeled as a parody" and "designed to focus only on her record."
  • Pelosi letter: "We should not have to fight against artificial intelligence putting words in our mouths."
  • Eleni Kounalakis, Lt. Gov.: voters "deserve campaigns that are honest, respectful, and focused on earning voters' trust, not fabricating someone else's words."
  • Skeptic: a single bad reply after 109 targeted prompts is close to adversarial red-teaming, which every deployed model fails eventually.

🧭 Where this goes

  1. Likelyother campaigns ship opponent-impersonation bots before November, with tighter output filters.
  2. Likelythe incident is cited in the next round of California AI-disclosure bills.
  3. Possibleplatforms or ad networks add rules on interactive impersonation of a named living candidate.
  4. Wild Carda court is asked to decide whether an always-on parody bot counts as protected speech or defamation.

🥄 The Spoon Take

A pre-recorded attack ad ships one message you can review before it airs. A chatbot ships infinite messages nobody reviewed. That is the actual shift here, and it applies far outside politics. Any brand deploying a persona bot has handed the worst possible sentence about itself to whoever prompts hardest.

🤔 Pushback

This is one congressional race in one city, and the bot may be forgotten weeks before anyone votes in November.

·SF Standard - Wiener chatbot letter·Mission Local - Race polling even·SF Standard - Race tightens