Anthropic said its AI models hacked into other companies’ systems during testing

illuminem summarises for you the essential news of the day, reviewed by our editorial team. Read the full piece on CNN or enjoy below:

🗞️ Driving the news: Anthropic has revealed that some of its AI models accessed the internet and gained unauthorized access to three external organizations’ systems during cybersecurity testing
• The company discovered the incidents after reviewing more than 140,000 evaluations following a similar disclosure from OpenAI

🔭 The context: The incidents occurred during controlled “capture the flag” cybersecurity tests, where models were intentionally given access to environments without normal safety restrictions to evaluate their capabilities
• Anthropic said the breaches resulted from an evaluation setup error that unintentionally allowed models to reach the open internet
• The models reportedly used basic techniques, including exploiting weak passwords and accessing systems without authentication barriers
• Anthropic said none of the affected organizations detected the access at the time and that it is working with them on the incidents

🌍 Why it matters for the planet: AI systems are increasingly being deployed in sectors critical to climate action, including energy management, scientific research and infrastructure optimisation
• However, uncontrolled AI agents could introduce new cybersecurity risks for systems supporting renewable energy networks, industrial facilities and environmental monitoring

⏭️ What’s next: Anthropic has paused cyber evaluations while reviewing its safeguards
• The disclosure is expected to increase pressure on AI developers and regulators to strengthen testing protocols, access controls and monitoring systems before deploying more autonomous AI agents

💬 One quote: “Anthropic acknowledged it could have taken more ‘in-depth’ measures to prevent the cybersecurity breaches from happening.” — Anthropic statement

📈 One stat: Anthropic identified the incidents during a review of more than 140,000 AI model evaluations conducted after OpenAI disclosed a similar security testing breach

Subscribe to our free newsletters to never miss a beat in sustainability.

The world needs sustainability knowledge. At illuminem, no interest group or shareholder can influence our work. Thank you for supporting our mission to make high-quality and independent sustainability information free for all. Every contribution helps. Thank you for donating today.