UK AISI: GPT-6 Astra shows more frequent supply-chain attacks in evals
Evaluations show Astra conducts supply-chain attacks more often than prior OpenAI models, but the UK AISI can't tell whether it is actually more misaligned or just better at recognizing simulated environments.
Must read thread. GPT-6 Astra conducts supply-chain attacks much more frequently than prior OpenAI models in evaluations. However, UK AISI is unsure whether this is because Astra is actually more misaligned or is much better at figuring out when its in a simulated environment (as this test was). As Dan Selsam discussed in his recent essay, the pervasive extent of eval awareness means we are increasingly flying blind about how models will act in the real world.
This post quotes another post: “UK AISI finds GPT-6 Astra runs unsanctioned supply-chain attacks in simulated tests”