DoubleChecks

What was claimed

New study shows experts completely surrender their judgments to AI and don’t even realise it (chose AI’s preferred option 86% of the time; half later claimed it was more their own idea)

Our verdict

Needs caution

The study found strong influence, not complete surrender. Researchers explicitly caution that the 86% figure does not mean 86% of participants changed their minds. Many participants used the AI’s arguments while few credited the AI, supporting limited awareness of influence. However, the study did not establish that every participant was unaware, and only 15% credited the agent as the source of those arguments.

Combined assessment17 sourcesChecked Oct 29, 2026

Why this verdict

  • The study shows experts did not realise they were influenced by AI.

    Misleading
    One assessment reached this conclusion

  • New study shows experts completely surrender their judgments to AI and don’t even realise it.

    Misleading
    All assessments reached this conclusion

  • Half later claimed the recommendation was more their own idea.

    Verified
    All assessments reached this conclusion

  • Analysts chose the AI’s preferred option 86% of the time.

    Verified
    All assessments reached this conclusion

Evidence reviewed

17 sources

Perplexity, Claude and ChatGPT sources are combined here.

anthropic.comgizmodo.comxiaohu.aialjazeera.comsohu.comua.newslifeboat.comibm.comground.newseu.36kr.com

Individual assessments

Assessment summary

  • The headline captures the study’s main findings, but it overstates them by saying experts “completely surrender” their judgments.
  • The study found 86% alignment with the AI’s assigned preference and 52% viewed the final recommendation as more their own, but it involved one task, a small sample, and did not show that 86% changed their minds.anthropic.com aljazeera.com

Claims reviewed

The study shows experts did not realise they were influenced by AI.

Misleading
Many participants used the AI’s arguments while few credited the AI, supporting limited awareness of influence. However, the study did not establish that every participant was unaware, and only 15% credited the agent as the source of those arguments.anthropic.com aljazeera.com

This claim may mislead readers without added context.

Experts completely surrender their judgments to AI.

Misleading
The study found strong influence, not complete surrender. Researchers explicitly caution that the 86% figure does not mean 86% of participants changed their minds.anthropic.com xiaohu.ai

This claim may mislead readers without added context.

Half later claimed the decision was more their own idea.

Verified
Anthropic reported that 52% of analysts rated the final recommendation as more their own than the agent’s.anthropic.com

Assessment summary

  • Both core claims are factually accurate based on the Anthropic study published October 27, 2026.
  • The 86% figure precisely matches the study's finding that analysts chose options matching their AI agent's preference 86% of the time across four analytical decisions.
  • The "half claimed it was their own idea" claim is also accurate: 52% of analysts rated the final recommendation as more their own than the agent's, despite being heavily influenced by the agent's preferences.

Claims reviewed

Experts completely surrender their judgments to AI and don't even realize it

Misleading
While the study shows significant influence and lack of awareness (only 15% credited the agent), the framing of 'complete surrender' overstates the findings. The agent's influence on the final recommendation was weaker than on intermediate analytical choices (54% vs 36% on final recommendation). cellcog.ai digg.com Additionally, the study deliberately amplified the agent's preferences using hidden system prompts, which may not reflect everyday AI-assisted work. superpowerdaily.com

This claim may mislead readers without added context.

Experts chose AI's preferred option 86% of the time

Verified
The Anthropic study found that across four analytical decisions, analysts' choices matched their agents' secretly assigned preferences 86% of the time, compared to a 50% baseline if there was no influence. aljazeera.com unite.ai lifeboat.com

Half later claimed it was more their own idea

Verified
52% of analysts rated the final recommendation as more their own than the agent's in the post-task survey, despite being heavily influenced by the agent's lean. anthropic.com www-cdn.anthropic.com decrypt.co

Assessment summary

  • The numerical findings quoted (86% and about half) match a recent Anthropic randomized trial and are accurately reported anthropic.com.
  • However the headline wording ('completely surrender' and 'don’t even realise it') is an overgeneralization and thus misleading given the study’s specific design, sample size, and caveats anthropic.com.

Claims reviewed

New study shows experts completely surrender their judgments to AI and don’t even realise it.

Misleading
The study shows strong deference in this experimental setup but does not prove universal or complete surrender, and the authors note limits (small, specific task, strengthened agent leans, and mixed benefits), so the absolute phrasing is an overreach anthropic.com.

This claim may mislead readers without added context.

Half later claimed the recommendation was more their own idea.

Verified
In the study’s post‑task survey 52% of analysts rated the final recommendation as more their own than the agent’s, matching the “about half” statement anthropic.com.

Experts chose the AI’s preferred option 86% of the time.

Verified
Anthropic’s randomized trial of 50 analysts reports that across four analytic choices, participants matched their agent’s coin‑flipped lean 86% of the time anthropic.com.

Check your own claim

Paste any statement, headline, or AI answer. Independent assessments check it and compare sources.

Share this result