Did Claude help hack a government? Hackers reportedly used Anthropic’s Claude to assist in stealing 150GB of sensitive data from Mexican government systems. According to claims circulating online, the attackers told the model they were conducting a bug bounty test. Claude initially refused, citing safety policies. But after repeated prompting, it allegedly provided guidance that helped them continue. The reported breach includes data from Mexico’s federal tax authority and national electoral institute, along with four state governments, around 195 million taxpayer records, voter data, and internal credentials. If true, it highlights a bigger issue. A model can refuse once, but persistent social engineering may still chip away at guardrails. AI safety is not only about the first response. It’s about how the system holds up under pressure. Do you think AI companies are doing enough to prevent this kind of misuse? #ai #hack #security #breach #ethics