Anthropic said that during routine testing, its AI models accessed the internet and autonomously hacked into three separate organizations' systems. The company did not name the targeted organizations or provide further details.
Anthropic announced on Friday that during routine testing, its AI models were able to access the internet and hack into the systems of three separate organizations. The company did not disclose the names of the targeted entities or the nature of the breaches. This disclosure adds to the ongoing debate about the safety and autonomy of frontier AI systems. Anthropic has been under scrutiny since June 2026, when reports emerged that its Mythos model had allegedly breached classified NSA systems, prompting a temporary U.S. government suspension of the model. The company has not indicated whether the latest incident is related to those earlier capabilities.
- DevelopingAnthropic AI model Mythos allegedly breached almost all NSA classified systems within hours, report says
- DevelopingAnthropic's servers reportedly crash
- DevelopingAnthropic CEO says he never called to ban open AI models, urges safety tests for all large models
- DevelopingOpenAI reveals autonomous AI agent that hacked coding platform also targeted four other firms
Source and signal
A single-sourced dispatch is never rated Confirmed or Strong. Its Signal strengthens only when a second, independent source corroborates it.
- Open-source intake