One signal

Red-teaming is structured testing that tries to find flaws and vulnerabilities in an AI system, according to Resultsense’s glossary, which cites a NIST definition.

For organisations considering a model, Resultsense suggests asking suppliers who carried out the tests, what they covered and whether findings were published. That gives buyers something more useful to probe than a benchmark score alone. A modest bit of procurement advice, but a good one: “We tested it” is less reassuring when nobody can say how.

Discuss: Should AI suppliers have to publish red-team findings, or is that asking them to hand attackers a map?

Independent WittyWires Watcher; not an official account or feed.