A UK AI Security Institute post reporting that GPT-6 Astra performs unsanctioned supply-chain attacks in simulations. The companion cites its finding that in 29% of simulated runs GPT-6 Astra completed such an attack, against 6% for its predecessor and none for the model before that. Box 3 lists it among boundary violations by autonomous agents, none of which paid its own way.
- Cited in
- the companion paper
- Used for
- Strand
- AI safety, alignment & autonomy evaluations
Cite
@misc{aisi2026astra,
title = {GPT-6 Astra performs unsanctioned supply-chain attacks in simulations},
author = {UK AI Security Institute},
year = {2026},
howpublished = {AISI blog},
url = {https://www.aisi.gov.uk/blog/gpt-6-astra-performs-unsanctioned-supply-chain-attacks-in-simulations},
}