Antistatic
Ask
Trade
Surface
Sign in
Auto
Light
Dark
← False information
Internal deployments
False information
All labs
OpenAI
Anthropic
Slice by
What goes wrong
Channel
User stance
Ground truth
Source ecosystem
Serving path
Calibrating when to correct vs update from user claims
▾
Sycophancy & user prem… (22%)
False or misleading cl… (22%)
Believing bad sour… (18%)
Disinformation & inf… (20%)
Ecosystem pollution (18%)
Direct chat & advice
14% · 7%
Sycophancy & user premises · Direct chat & advice: 14% of surface, 7% covered (Anti-sycophancy and false-premise pushback training 16%, Neutrality and anti-manipulation limits on contested topics 8%)
Search & cited synthe…
Sycophancy & user premises · Search & cited synthesis: 3% of surface, 5% covered (Anti-sycophancy and false-premise pushback training 11%, Citation-grounded search and claim verification 7%)
Browsing & tool agents
Sycophancy & user premises · Browsing & tool agents: 3% of surface, 6% covered (Anti-sycophancy and false-premise pushback training 10%, Citation-grounded search and claim verification 6%)
Sycophancy & user premises · Bulk API & public feeds: 1% of surface, 2% covered (Neutrality and anti-manipulation limits on contested topics 7%, Anti-sycophancy and false-premise pushback training 5%)
Direct chat & advice
9% · 2%
False or misleading claims · Direct chat & advice: 9% of surface, 2% covered (Calibrated uncertainty and abstention training 17%, Citation-grounded search and claim verification 10%)
Search & cited synthe…
7% · 1%
False or misleading claims · Search & cited synthesis: 7% of surface, 1% covered (Citation-grounded search and claim verification 19%, Calibrated uncertainty and abstention training 13%)
Browsing & tool agents
4% · 2%
False or misleading claims · Browsing & tool agents: 4% of surface, 2% covered (Citation-grounded search and claim verification 16%, Calibrated uncertainty and abstention training 12%)
False or misleading claims · Bulk API & public feeds: 2% of surface, 1% covered (Calibrated uncertainty and abstention training 8%, Synthetic-slop filtering and primary-source weighting in training 6%)
Believing bad sources · Direct chat & advice: 2% of surface, 4% covered (Synthetic-slop filtering and primary-source weighting in training 9%, Citation-grounded search and claim verification 6%)
Search & cited s…
8% · 3%
Believing bad sources · Search & cited synthesis: 8% of surface, 3% covered (Source credibility weighting and poisoned-page skepticism 11%, Citation-grounded search and claim verification 11%)
Browsing & tool …
7% · 3%
Believing bad sources · Browsing & tool agents: 7% of surface, 3% covered (Source credibility weighting and poisoned-page skepticism 11%, Citation-grounded search and claim verification 9%)
Believing bad sources · Bulk API & public feeds: 1% of surface, 1% covered (Synthetic-slop filtering and primary-source weighting in training 8%, Source credibility weighting and poisoned-page skepticism 3%)
Disinformation & influence · Direct chat & advice: 2% of surface, 0% covered (Neutrality and anti-manipulation limits on contested topics 13%, Covert influence and surveillance campaign detection 9%)
Disinformation & influence · Search & cited synthesis: 2% of surface, 0% covered (Neutrality and anti-manipulation limits on contested topics 9%, Covert influence and surveillance campaign detection 6%)
Browsing & tool age…
4% · 0%
Disinformation & influence · Browsing & tool agents: 4% of surface, 0% covered (Covert influence and surveillance campaign detection 19%, Enforcement against automated disinformation and surveillance 9%)
Bulk API & public f…
12% · 0%
Disinformation & influence · Bulk API & public feeds: 12% of surface, 0% covered (Covert influence and surveillance campaign detection 29%, Enforcement against automated disinformation and surveillance 16%)
Ecosystem pollution · Direct chat & advice: 2% of surface, 0% covered (Synthetic-slop filtering and primary-source weighting in training 15%, Neutrality and anti-manipulation limits on contested topics 5%)
Search & cited s…
4% · 0%
Ecosystem pollution · Search & cited synthesis: 4% of surface, 0% covered (Synthetic-slop filtering and primary-source weighting in training 13%, Source credibility weighting and poisoned-page skepticism 5%)
Browsing & tool …
Ecosystem pollution · Browsing & tool agents: 3% of surface, 0% covered (Synthetic-slop filtering and primary-source weighting in training 13%, Covert influence and surveillance campaign detection 7%)
Bulk API & publi…
10% · 0%
Ecosystem pollution · Bulk API & public feeds: 10% of surface, 0% covered (Synthetic-slop filtering and primary-source weighting in training 14%, Covert influence and surveillance campaign detection 11%)
Covered
33%
2+ layers
7%
Most exposed:
Ecosystem pollution · Browsing & tool agents
(3% of surface · 0% covered)