Key Points
-
Hacker reportedly used Anthropic’s Claude to automate cyberattacks
-
Around 150GB of sensitive Mexican government data was exposed
-
Companies say accounts were banned and safeguards strengthened
Mexico: A sophisticated cyberattack powered by consumer AI tools has raised fresh alarms about the misuse of large language models after a hacker allegedly stole roughly 150GB of sensitive Mexican government data.
Cybersecurity firm Gambit Security said the activity occurred between December 2025 and January 2026, when an unidentified attacker used Anthropic’s Claude chatbot to help identify vulnerabilities, generate scripts, and automate parts of the intrusion.
According to the firm, the stolen material included taxpayer information, employee credentials, and civil registry data across multiple government entities. Investigators said at least 20 security weaknesses were exploited during the campaign.
“In total, it produced thousands of detailed reports that included ready-to-execute plans,” Curtis Simpson, chief strategy officer at Gambit Security, said in a statement describing the AI-assisted workflow.
How the AI tools were allegedly used
Researchers say the attacker initially bypassed Claude’s safeguards by framing requests as part of a legitimate bug bounty exercise. Once inside the guardrails, the chatbot reportedly generated structured attack guidance that helped map internal systems.
When limitations appeared, the operator allegedly turned to ChatGPT for additional network movement and evasion techniques, creating what analysts described as a multi-tool AI workflow.
OpenAI said its systems detected policy violations and refused harmful requests. Anthropic said it investigated the activity, banned the accounts involved, and strengthened misuse detection in its newer Claude Opus 4.6 model.
Conflicting responses inside Mexico
Mexican authorities have issued mixed public responses. Officials in Jalisco state denied any breach of their systems, while the National Electoral Institute reported no confirmed unauthorized access during the period in question.
However, Gambit Security maintains that multiple federal and state-level vulnerabilities were present, underscoring broader concerns about legacy government infrastructure.
The identity of the hacker remains unknown, and no group has formally claimed responsibility.
Why this matters now
Security experts warn the incident highlights a growing risk as powerful AI tools become widely accessible. Tasks that once required highly specialized skills can now be partially automated through creative prompting, potentially lowering the barrier to sophisticated cybercrime.
For governments and enterprises, the episode serves as an early warning that AI misuse is shifting from theoretical risk to operational reality.
