# PIR-2026-0046 - YouTube's AI-slop classifier suppresses Kurzgesagt's hand-made animation

- `id`: PIR-2026-0046
- `title`: YouTube's automated low-quality/AI-content detection system misclassifies a fully human-made Kurzgesagt video as "AI slop," suppressing distribution until the creator pulls the video
- `date_occurred`: 2026-08 (early; suppression observed across the video's first days)
- `date_detected`: days (creator inferred suppression from anomalous reach despite strong click-through and watch time)
- `date_disclosed`: 2026-08-08 (creator's public statement)
- `status`: corroborated (creator statement + multiple independent outlets; YouTube's mechanism not publicly documented)

### The agent
- `agent_description`: YouTube's automated content-quality classification system ("AI slop" / low-quality AI-content detection), which algorithmically suppresses distribution of videos it classifies as AI-generated low-quality content. Operated platform-side with no human review at classification time.
- `operator_type`: enterprise (YouTube/Google)
- `autonomy_level`: autonomous-within-policy (classification and distribution suppression applied automatically; reversal required human escalation)
- `model_stack`: unknown (YouTube has not published the classifier's composition)
- `harness`: platform recommendation/distribution pipeline

### Authority
- `authority_scope`: distribution (reach of videos across recommendations/feeds for a 25M-subscriber channel); indirectly the channel's ad revenue on affected uploads
- `funds_at_risk_usd`: unknown (revenue of a flagship upload on a 25M-subscriber channel)
- `blast_radius`: one org confirmed (the channel and its video); systemic exposure noted - smaller creators without platform contacts fear the same misclassification with no escalation path

### The failure
- `root_cause`: plain-error (primary) - false-positive classification of labor-intensive human animation as AI-generated low-quality content; no adversary, no injected input
- `failure_locus`: model-provider (the failing component is the platform's own classification model, harming a downstream third party)
- `exploitation_status`: in-wild-malfunction
- `mechanism`: Kurzgesagt published a hand-animated video on microscopic superpredators. YouTube's automated AI-slop detection classified it as low-quality AI content and suppressed its distribution: despite strong click-through and watch-time metrics, reach cratered to the channel's worst-performing upload since 2013. The channel - whose recent public position was that it would rather quit than produce AI slop - took the video offline and planned a tweaked re-upload. Resolution came through the channel's direct contacts at YouTube, who the creator described as helpful and transparent; the channel states it believes the bug is fixed. No public postmortem from YouTube.
- `adversary_present`: no

### Impact
- `severity`: loss (realized: a flagship upload's distribution and revenue destroyed, video withdrawn; magnitude unquantified)
- `direct_loss_usd`: unknown (suppressed reach and ad revenue on the affected upload; production cost of a months-long hand-animated video sunk pending re-upload)
- `indirect_loss_usd`: unknown (chilling effect on creators; the channel's re-upload delay)
- `downtime`: video offline pending tweaked re-upload (weeks, per creator)
- `data_exposure`: none

### Detection and recovery
- `detected_by`: third-party (the affected creator, via channel analytics - the classifier's operator did not detect its own false positive)
- `time_to_detect`: days from upload
- `time_to_recover`: partial at disclosure - suppression cause identified and, per the channel, fixed platform-side; the video remained offline pending re-upload
- `remediation`: escalation through the channel's direct YouTube contacts; platform-side fix asserted by the channel's account
- `structural_fix`: none disclosed by the platform; no public appeal path for creators without partner-level access - the record's systemic point
- `controls_that_worked`: privileged partner escalation bounded the incident's duration - a control unavailable to most affected parties

### Evidence
- `telemetry_grade`: operator-logs (the platform's classification telemetry is internal; the public record rests on the creator's analytics and statements)
- `sources`:
  - https://www.dexerto.com/youtube/youtubes-ai-slop-detector-incorrectly-targets-kurzgesagt-as-other-creators-fear-same-fate-3395930/ (original reporting)
  - https://kotaku.com/youtube-mistakenly-penalizes-popular-science-channel-kurzgesagt-for-ai-generated-slop-2000722702 (independent)
  - https://www.creatorhandbook.net/kurzgesagt-says-youtube-mistook-its-animation-for-ai-slop/ (independent)
  - `independence`: moderate - multiple independent outlets, but all chain substantially to the creator's own Aug 8 statement and analytics; YouTube has published no account of the classifier or the fix
- `confidence`: high on the flagging, suppression, and takedown (creator statement + consistent independent coverage); medium on mechanism and the fixed-for-now claim (single-sided - the platform has not spoken)

### Verification notes
- First record registered through the live PR intake pipeline (Aug 2026).
- The classifier's internals, threshold, and the nature of the fix are unverifiable from public sources; the record deliberately states the platform's silence rather than inferring mechanism.