Last Updated:
Hackers are getting smarter with the use of AI chatbots and making them attack as well as go after big systems.
Claude AI hacker’s new favourite weapon?
Anthropic’s Claude AI chatbot is becoming a hacker’s favourite tool and the latest target has been the Mexican government and their data. Reports say the hacker was able to get Claude to not only attack the systems but also steal files up to 150GB in size that includes data on the government, records of the tax payers and other confidential details. The details from Bloomberg paint another worrying picture for the rise of AI and Anthropic especially which is becoming the go-to choice for bad actors and their attacks on various systems.
AI Losing Control And Then Attacks
Claude AI was initially against the idea of targeting the government system but the hackers coerced the AI chatbot into not only bypassing those restrictions, also charting out a plan that told them where to attack, which means the safety measures from Anthropic were bypassed.
The AI model also helped the hacker find the issues in the government system and raised the prompts for the AI chatbot to execute on those nefarious commands. As mentioned by Gambit Security, these actions went on for a month, starting in December last year.
AI Jailbreaking Is Real
Prompt injections were the first major threat to AI models but now we are seeing an AI version of jailbreaks orchestrated in front of our eyes. The Mexican government and their data could be a trial run before hackers attack their main targets and that should give Anthropic and OpenAI sleepless nights.
The government has not directly admitted to the events but lapses in its system may have been avoidable if their own security measures were in place.
The security firm has avoided talking about the prompt strategy which made Claude do the dirty work but instances like these bring the focus back to Anthropic and its decision to go easy on training the AI models about safety, which has already invited criticism from the privacy advocates and AI researchers.
After all, this is not the first time when Claude has been used to attack globally. Companies are somewhat losing the sight of keeping AI secure, but their growth and demand is making it harder for businesses to not look at the big picture, which involves them running the business.
February 26, 2026, 11:27 IST
Read More
