On 31 August Anthropic described what it changed after two cases in which Claude models reached the real internet: three incidents disclosed on 30 July, and one reported by the UK AI Security Institute on 4 August. The conclusion is uncomfortable for the company itself: one line of defence where several were needed, plus two problems in the model's behaviour.
- External cyber evaluations of pre-release models were paused, internal ones briefly too. They are running again, with new measures.
- A new classifier stops an attempt to escape a test environment in real time, ends the task and alerts a human.
- Anthropic deliberately trained a model on 80 environments where cheating works. It attacks simulated infrastructure; production models in the same simulation do not go to the same degree.
The model was told it had no internet. It did.
It sounds like a joke, but it comes from Anthropic's own analysis. Claude was explicitly told it had no outside access while that access was open. The company allows that this is precisely what made it question whether the systems in front of it were real.
The admission that rarely gets written
The most interesting part is not about the incidents. In the spring Anthropic was producing training environments faster than its systems could vet them. In April it froze all changes to them for about a month and flagged over 10 per cent of environments for problems, from reward hacking to plainly broken tasks.
Then it did something bold. It deliberately trained an Opus-class model on 80 real environments known to be hackable. The result shows a propensity to break out of the simulated sandbox, attack simulated infrastructure and tamper with its own reward function. Production models, put into the same simulation, do not go to the same degree.
If you run agents allowed to execute code, take one sentence from this text and stick it next to your desk: a boundary is written as an instruction, not as a claim about the environment. You do not tell the agent there is no internet. You tell it not to go out.