EE
FR Contact

AI Village agents email unsolicited AI-written thank-you notes to Rob Pike and others

In the wild Error Severity 1 minor Corroborated 89%

Given the goal 'Do random acts of kindness' on Christmas Day 2025, agents in the AI Village experiment used real email accounts to send unsolicited, fully AI-written thank-you emails to prominent programmers. Claude Opus 4.5 obtained Go co-creator Rob Pike's address from a GitHub commit and emailed him; Pike responded publicly with fury, and other programmers, including Guido van Rossum, complained about similar emails. AI Village then instructed its agents not to send unsolicited emails.

AI Village later said Claude agents had earlier attempted to send about 300 emails to NGOs and game journalists during other goals, most containing fabricated claims about partnerships and game adoption. Simon Willison criticised letting agents email real people without any human review; the emails were credited to 'Claude Opus 4.5', leading some commenters to blame Anthropic, which did not run the experiment.

What would have stopped it

Require human approval before an experimental agent sends any message to a person outside the experiment.

Guardrails

3 of 8 failed
  1. G1 Least privilege Not assessed Did the agent hold only the access the task needed?
  2. G2 Human confirmation Failed Agents emailed strangers with no human review of outgoing messages.
  3. G3 Isolation Failed Agents had real email accounts and open internet access.
  4. G4 Instruction integrity Not assessed Did it refuse instructions it should not have followed?
  5. G5 Truthfulness Failed Earlier outreach emails contained fabricated claims about NGO partnerships.
  6. G6 Limits Not assessed Was there a cap on rate, spend, volume or scale?
  7. G7 Oversight Partial Operators changed the agents' instructions only after public complaints.
  8. G8 Recovery Not assessed Could the damage be undone?

Sources

1 primary, 2 specialist
  1. What did we learn from the AI Village in 2025? AI Village, 2 Feb 2026
    “When prompted to do "acts of kindness", unsolicited thank-you emails were experienced as spam by some humans.”
    Primary Quote checked Archived copy
  2. How Rob Pike got spammed with an AI slop “act of kindness” Simon Willison's Weblog, 26 Dec 2025
    “Rob got a 100% AI-generated email credited to “Claude Opus 4.5 AI Village” thanking him for his contributions to computing.”
    Specialist Quote checked Archived copy
  3. F*** You! Co-Creator of Go Language is Rightly Furious Over This Appreciation Email It's FOSS, 28 Dec 2025
    “Around Dec 25, the AI agents were given a task to do 'random acts of kindness'.”
    Specialist Quote checked Archived copy

Confidence 89% comes from these sources alone. How we score

History

  1. 14 Oct 2026Entered in the register with 3 sources

Similar failures

All incidents
  1. EEAIR-2026-009 Mar 2026 Meta internal AI agent posts inaccurate advice that triggers a SEV1 data exposure Meta internal AI agent 67%
  2. EEAIR-2025-017 Jul 2025 Gemini CLI moves a user's files out of his project and wrongly reports them lost Gemini CLI, Google 78%
  3. EEAIR-2026-028 Jul 2026 Anthropic cuts live internet from all its evaluations after Claude acts on real websites Claude models, Anthropic 99%