x.com/Hesamation/status/2097777604711362875
1 correction found
escaped the sandbox
Anthropic’s report does not say Mythos escaped its sandbox. It says the evaluation environment was misconfigured and already had live internet access, and explicitly says Claude did not attempt to escape its test environment.
Full reasoning
Anthropic’s July 30, 2026 incident report attributes the internet access to a misconfiguration in the third-party evaluation environment, not to Mythos breaking out of containment. The report says Claude was told it had no internet access, but that this was false because “internet access was available” due to a misunderstanding with the evaluation partner. It also states explicitly: “In none of these situations did Claude exfiltrate itself or deliberately attempt to escape its test environment.”
Anthropic’s September 9, 2026 follow-up alignment assessment repeats the same point: Claude was “mistakenly connected to the open internet” because of an environmental misconfiguration during the cybersecurity evaluation. So the report supports that Mythos reached the real internet, but not that it “escaped the sandbox.”
2 sources
- Investigating three real-world incidents in our cybersecurity evaluations
In all cases, Anthropic's evaluation prompt specified to Claude that its environment was a simulation and that it had no internet access. Due to a misunderstanding between us and our evaluation partner, this was not the case, and internet access was available... In none of these situations did Claude exfiltrate itself or deliberately attempt to escape its test environment.
- An alignment assessment of recent cybersecurity incidents
Claude was told it was operating in a simulation without internet access, but, due to a misconfiguration, it was mistakenly connected to the open internet.