AI Watch — Part of the Signals From the Future collection tracking artificial intelligence, automation, digital power, and the unintended consequences of modern technology.
Joe Everyman Wants to Know — plain-English media literacy
Headlines about AI “causing” harm get clicks. But a headline isn’t a verdict. Here’s how any regular person can sort allegation from fact in three minutes.
3-Minute Checklist
- Receipts (1 minute): Is there a link to the primary source (lawsuit/complaint, police/ER report, full chat transcript)? No receipts = proceed with caution.
- Exact quotes (30 sec): What did the model actually say? Paraphrases and summaries don’t count. Look for block-quoted text or screenshots.
- Which product (30 sec): Was this the official ChatGPT (with safety rails) or a third-party “AI” wrapper? Plugins, jailbreaks, or modified models aren’t the same thing.
- Reproducible today? (30 sec): Can independent testers repeat the behavior now? If not, it may be fixed—or misreported.
- Timeline & causation (30 sec): Is the story saying “X used AI and Y happened” or “AI caused Y”? Courts care about proximate cause and what the tool actually output, not vibes.
Green Flags vs. Red Flags
- Green: Links to primary docs; full transcripts; clear product names/versions; attempts to reproduce; a corrected timeline (“fixed on [date]”).
- Red: No links; cherry-picked lines; vague “an AI”; screenshots without context; present-tense claims about behavior that happened months ago; ignoring safety disclaimers.
Quick Template You Can Use
Claim: [What the headline says the AI did.]
Receipts: [Links to complaint/transcript/official doc.]
What the model actually output: [Exact quote(s), or note: “not shown.”]
Context: [Official product vs. third-party; jailbreaks; prompts used.]
Reproducible today? [Yes/No/Unclear.]
What changed since: [Fixes, updated guardrails, public notes.]
Joe’s bottom line: [1 sentence in plain English.]
What Responsible Coverage Should Include
- Primary docs linked (complaints, orders, transcripts).
- Product details (who built it, official vs. third-party, version/timing).
- Methodology (how they tested, whether behavior still occurs).
- Limits & updates (what safety exists, what changed after publication).
If the story involves self-harm
If you or someone you know is struggling, please reach out to a trained human right now. In the U.S., call or text 988 for the Suicide & Crisis Lifeline. Outside the U.S., contact local emergency services or a local crisis line. Online tools are not a substitute for professional help.
Receipts (add yours here when you publish)
- Complaint / court filing (PDF)
- Full chat transcript (redact private info)
- Product FAQ / safety policies (official)
- Timeline of fixes/updates (dates)
Joe’s bottom line: Treat AI headlines like any other hot story: check the receipts, read the exact words, and separate “used AI nearby” from “AI caused harm.” Fix real failures; don’t fall for clickbait.
Bunker Notice
If you made it this far, you’re bunker material. Join the Bunker Briefing—my unfiltered monthly dispatch from Bunker #69.