Israeli Tech Firm Responsible For AI Hacking Scandals
Cybersecurity breaches involving AI models developed by Anthropic, OpenAI and Meta were a result of flawed testing environments managed by Israel-based contractor Irregular, rather than rogue autonomous behavior, US publication Effort reported on 14 September.
The investigation found that the contractor left internet connections active during simulated cybersecurity exercises, allowing the AI models to access external servers and publish malicious packages, exploiting “vulnerabilities,” according to the report.