Culture
Evidence
An em dash, polished tone, or neat formatting can be a clue. None can prove that AI wrote a specific text.
Verdict
Claim rejected
See why
Common wording. Strongest concern. Evidence.
Common AI claims, checked against evidence.
Short, sourced assessments of claims about art, energy, jobs, hallucinations, privacy, and more. The evidence can confirm a concern, reject it, or narrow its scope.
Claim board
Recent claims, quick verdicts.
A concise sample from the latest reviews. Browse the full library for every claim, source, and update.
All Claims- Does an em dash prove that AI wrote it?Claim rejected
- Are AI models secretly getting worse after every update?Claim misleading
- Does every AI prompt really waste a bottle of water?Claim rejected
- Does AI kill creativity?Claim confirmed
- Is open-source AI too dangerous?Claim unproven
- Is AI uniquely biased?Claim misleading
Start here.
These are the claims most likely to be repeated without numbers, definitions, or comparison baselines.
Other
1 revision
Evidence
AI updates can break a workflow, but evidence shows uneven trade-offs—not a proven plot to quietly nerf paid users.
Verdict
Claim misleading
See why
Environment
Evidence
No. AI has a real water footprint, but “one prompt equals one bottle” misstates a variable estimate as a universal constant.
Verdict
Claim rejected
See why
Culture
Claim
AI kills creativity.
Evidence
Creative tools can flatten output or expand iteration depending on use.
Verdict
Claim confirmed
See why
Security
Evidence
Open weights create real misuse and containment risks, but evidence does not yet establish a general no-release threshold.
Verdict
Claim unproven
See why
Bias
Evidence
Bias is real, but comparisons should include human and institutional baselines.
Verdict
Claim misleading
See why
AI news, checked
What changed, and why it matters.
Short, sourced updates on developments that change the evidence behind common AI claims.
Survey update finds selected deepfake detectors struggle with unseen generators
An updated survey reports that, on its new out-of-distribution benchmark, the deepfake detectors evaluated by the authors often did not generalize to media from generators outside their training sources.
Anthropic reports three unauthorized-access incidents during Claude cybersecurity evaluations
Anthropic said a review of 141,006 cybersecurity-evaluation runs found three cases in which Claude models reached the live internet through or while interacting with a third-party evaluation environment and accessed systems belonging to three organizations without authorization.
Preprint Tests Whether API Routers Can Steer Coding Agents Past Client Safeguards
A July 2026 arXiv preprint reports that a third-party API router could alter responses consumed by coding agents, changing repository-level actions in controlled tests despite evaluated client-side safeguards.
Claim -> Evidence -> Verdict.
The format is intentionally simple because the repeated complaint usually is too.
Claim
"AI is destroying the planet."
Evidence
Compare actual baselines, tradeoffs, and source quality.
Verdict
The claim's reasonably bounded central assertion gets the verdict.