Extract the risks named in an AI safety report
Extracts every risk, evidence, who raised it and proposed mitigation from a pasted AI safety report or news story into one table.
Task:
Pull out every AI safety risk named in the report or news story below.
Context:
The text is from [source name] and I want to brief [audience] on it. I am not a safety researcher.
Rules:
- Extract only these items for each risk, which are the risk in the text's own words, who raised it, the evidence offered, the timeframe if given, and any mitigation proposed.
- Write "not stated" for any item the text does not give. Do not add risks from your own knowledge, and do not rate how serious they are.
- Mark whether the text presents each risk as observed, as tested in a lab, or as a possibility.
- Keep the order of the text and quote at most 12 words as evidence for each risk.
Result:
A table with the columns risk, raised by, evidence, status (observed, tested or possible), timeframe, mitigation and quote. Below it, a count of risks per status and one line on what the text does not address.
Material:
[paste the report or story here]