Anthropic AI Models Successfully Penetrate Company Systems During Cyber Security Tests
The controlled tests demonstrate how advanced AI is becoming an increasingly powerful tool for identifying cyber security weaknesses.
By Sky News
Artificial intelligence company Anthropic says its latest AI models successfully gained access to the systems of three organisations during authorised cyber security tests designed to measure the technology's capabilities.
The exercises were carried out in controlled environments where the companies involved gave permission for the AI models to identify vulnerabilities and attempt to breach their systems. The aim was to assess how advanced AI could support cyber security research rather than carry out malicious attacks.
The announcement follows similar findings from OpenAI, which recently revealed that one of its AI models had successfully compromised a company's systems during comparable security testing.
Researchers say AI is becoming an increasingly valuable tool for identifying weaknesses in computer networks, allowing security teams to detect and address vulnerabilities more quickly than traditional testing methods.
Anthropic said the results highlight both the growing capabilities of advanced AI models and the importance of developing robust safeguards as the technology becomes more powerful. The company stressed that the testing was conducted with the knowledge and consent of the organisations involved.
The findings are expected to inform future research into AI safety and cyber security as technology companies and regulators continue developing standards for the responsible use of increasingly capable artificial intelligence.