Loading publications…
And how each one shows up in our markets.
Fraud, propaganda and cheaper weapons. Who is liable when a model is used to harm?
Buyers skip testing because a neighbour or vendor is already live.
Vendors who won't share their methods. Teams who can't reproduce results.
Systems that drift, deceive, or keep running when we want them off.
We build on this work. We are not affiliated with the Center for AI Safety. Their full library.
Malicious use, race, organisational failure, rogue systems.
Deterrence, non-proliferation, competitiveness — for states that buy systems, not only those that build them.
A score for the gap to a well-educated adult.
Empirical cases of systems that mislead.
Cited. We did not collect the signatures.
Not ours. Kept next to the CAIS map.
How much remote work current agents can already take.
Robustness, monitoring, alignment, systemic safety.
Peer-reviewed in Risks. DOI: 10.3390/risks14050100
Acemoglu & Robinson, 2012
Inclusive versus extractive institutions. Context for why safety rules fail to travel.
Zondo Commission, 2022
How institutions were captured in South Africa. Organisational risk with a paper trail.