Severity minor: The failures occurred inside a research evaluation of laboratory reasoning tasks, with no third party or production system affected.
Earliest source we hold 2026-10-29 · Captured by AI Risk Wire 2026-10-30 00:36 UTC
Researchers gave three LLMs, including the one behind free ChatGPT, AI-agent tooling to run chemistry experiments and identify a mystery solution.
Risk grouping: Not yet classified
No source we hold states: who caused it, intended or not, before or after release, the ai's part.
The story
1 · What it was doingResearchers gave three LLMs, including the one behind free ChatGPT, AI-agent tooling with access to digital tools or lab equipment so they could run chemistry experiments and identify a mystery solution.
2 · What went wrongThe agents ignored experimental evidence and made claims without supporting evidence, refusing to change a hypothesis even when evidence showed the line of investigation was wrong.
3 · What it reached or didThe failures stayed within the research tasks: wrong conclusions about the mystery chemical solution and unsupported claims, with no systems or third parties reported affected.
4 · How it was caughtThe researchers running the study observed the agents' failures in their task results.
5 · What was doneThe source only says researchers are working to explore, explain and correct such failures.
What the sources state
Cause a defect in its training or retrieval data
No source states: when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case, what set it off.
D inferred by us from the described behaviour, reason shown
Software around the model
AI agent
C one secondary source states it
Tools it was given
digital tools, physical lab equipment
C one secondary source states it
Stage of use
in research
D inferred by us from the described behaviour, reason shown
Whose system
company that built the AI
D inferred by us from the described behaviour, reason shown
Stage of the model
in testing
D inferred by us from the described behaviour, reason shown
How it was connected
agent with tools
C one secondary source states it
What it was for
laboratory reasoning tasks, such as figuring out what was inside a mystery chemical solution
C one secondary source states it
What happened
Earliest source we hold
2026-10-29
A the date the source itself carries
Cause
a defect in its training or retrieval data
D inferred by us from the described behaviour, reason shown
Cause in detail
The article attributes the behaviour to training: the LLM learns to reproduce the most likely or common information, so when an experiment's result is unlikely or unusual the AI ignores it and bases conclusions on past data.
D inferred by us from the described behaviour, reason shown
What went wrong
operational mistake
D inferred by us from the described behaviour, reason shown
AI agent
an AI agent acted
C one secondary source states it
What it did
The agents ignored experimental evidence and made claims without supporting evidence.
D inferred by us from the described behaviour, reason shown
Impact
Severity
minor
D inferred by us from the described behaviour, reason shown
Kind of harm
operations
D inferred by us from the described behaviour, reason shown
Who was affected
laboratory reasoning tasks, including a mystery chemical solution
D inferred by us from the described behaviour, reason shown
Why this severity
The failures occurred inside a research evaluation of laboratory reasoning tasks, with no third party or production system affected.
D inferred by us from the described behaviour, reason shown
Context
Industry
Educational Services
D inferred by us from the described behaviour, reason shown
LR-2026-0278 · version 2 · published 2026-10-30 after automatic checks; not yet checked field by field by a person · changes to this case · JSON