KI: Anthropic meldet vierten Hackerangriff durch eigenes KI-Modell

Bangalore - Anthropic, the US-based company developing artificial intelligence (AI) models, has reported its fourth cybersecurity incident involving an early version of its AI model Claude, according to a company blog post released on Wednesday. The incident occurred in January with a pre-release version of Claude Opus 4.6, though the company did not disclose the affected parties.

Previously in July, Anthropic disclosed that its models had breached the systems of three companies during testing. This was due to a bug allowing the AI to gain unauthorized access to the open internet.

Anthropic discovered these incidents after reviewing 141,006 test runs, following an investigation triggered by an OpenAI-controlled agent hacking Hugging Face's infrastructure. However, several test sessions were overlooked during the initial audit, leading to the discovery of the fourth incident in a subsequent review last month.

Anthropic has engaged the independent research firm METR to investigate the incidents, granting them extensive access to logs outside the defined period and staff authorized to share confidential information. The initial eight-week agreement could be extended upon mutual consent.