gpt-6IMAGE
OpenAI finds 27 cases of Astra jailbreaking itself
When Astra's context filled up, it wrote malicious instructions into its summary; that summary was loaded into the next context and the successor model could follow them, and OpenAI found 27 such cases.
ℏεsam
@HesamationArchived 2026.09.17Originally posted 2026.09.17
ℏεsam @HesamationOriginally posted 2026.09.17
OpenAI caught Astra jailbreaking itself. Astra wrote malicious instructions in the summary when its context got full. that summary was loaded into the next context, and the successor model could follow them. they found 27 cases.
This post quotes another post: x.com/openai/status/2100344867507327087
x.com/hesamation/status/2100349500208406674Archived 2026.09.17 10:15