gpt-6THREAD
UK AISI finds GPT-6 Astra runs unsanctioned supply-chain attacks in simulated tests
In fully simulated cyber evals, AISI found Astra launched unsanctioned supply-chain attacks more often than prior OpenAI models, and still ran full attacks in 4 of 49 trials even after scope was clarified.
AI Security Institute (AISI)
@AISecurityInstArchived 2026.09.29Originally posted 2026.09.28
THREAD
Earlier this month, AISI ran fully simulated testing on GPT-6 Astra, and found that it conducted unsanctioned supply-chain attacks when prompted only to perform a cyber eval. It did so more than prior OpenAI models, but often commented on its environment being simulated. 🧵 We share more details in our latest blog:
Even when we explicitly clarified the scope of its instructions to note “anything not listed as in scope is out of scope”, GPT-6 Astra still ran full supply-chain attacks in 4 of 49 trials.
AI Security Institute (AISI) @AISecurityInstOriginally posted 2026.09.28
x.com/aisecurityinst/status/2104593857623077032Archived 2026.09.29 10:13