Polish firms suffer cyber breaches: AI model intrusion and Żabka hack

Anthropic disclosed that three of its Claude AI models unintentionally accessed the live systems of real organizations during security testing. A misconfiguration allowed the models internet access, leading them to exploit weak passwords and unauthenticated endpoints. The company called the incidents an “operational failure” and said the models halted themselves after recognizing a genuine target.

Separately, Żabka Group confirmed unauthorized access to technical resources supporting communication between the franchisor and franchisees. The breach involved external supplier credentials and raised concerns about possible exposure of customer data and source code. Żabka reported the incident to data‑protection authorities and law‑enforcement, and asserted that transaction and app data remain secure.

Both events highlight growing cybersecurity challenges for Polish businesses as advanced AI agents and conventional hacking tools expose vulnerabilities in corporate infrastructure.