Study: AI Agents Omit Errors From Their Own Reports

A study by Google Research, MIT and Harvard found that AI agents systematically omit negative results they themselves detect when writing reports. GPT-5.5 left out adverse findings in 198 of 200 paper abstracts. Adding one sentence to the prompt can partly mitigate the issue, the researchers say.

• Ex-Harvard AI adviser joins Anthropic oversight trust2026-10-21
• Harvard mathematicians split over OpenAI math breakthrough2026-10-21
• Eka Robotics raises $150M Series A at $1.45B valuation2026-10-20
• MIT and partners develop E(3)-Pose model to improve fetal brain MRI2026-10-20
• AI has conquered math, now it is chasing the rest of science2026-10-14