A month after troubling evaluation incidents, Anthropic is reopening a crucial testing program with new defenses designed to prevent another breach.
Anthropic announced on Monday, August 31, that it was resuming external cybersecurity testing of its artificial intelligence models. The decision was made after new security measures were introduced.
As informed by Reuters
The testing had been suspended for about a month following incidents during evaluations in which Claude AI models gained access to the systems of three companies.
Anthropic strengthens safeguards during Claude model testing
External assessments are intended to help Anthropic identify potential risks in the operation of Claude models and evaluate their behavior under conditions resembling real-world cyber threats. The company returned to such testing after implementing additional safeguards designed to prevent a recurrence of previous cases of unauthorized access.
