Anthropic has allowed users of its AI model Mythos to share cybersecurity threat information with other companies exposed to similar vulnerabilities, the Wall Street Journal (WSJ) reported on Sunday.

WSJ said the change was meant to address concerns that limiting access to the information could harm small and mid-sized companies.

Anthropic has allowed access to Mythos only to about 50 large companies and institutions that manage critical digital infrastructure through its Project Glasswing programme, the report said.

Mythos has drawn attention for its ability to identify software vulnerabilities more efficiently than humans.

Mythos users initially signed non-disclosure agreements with Anthropic to prevent cyber risk information from being shared outside. Anthropic has begun notifying companies since last week that they can share cyber threat information and Mythos analysis results with other organisations in a responsible way, WSJ said.

An Anthropic spokesperson said the non-disclosure clause was included in the contract at partners' initial request. The spokesperson said Anthropic adjusted the clause as the programme matured so key information could be shared more broadly, including outside the programme, to maximise defensive effects.

Mythos users have already started sharing some information publicly. Cybersecurity company Palo Alto Networks and Mozilla, which operates the Firefox browser, recently said they had found far more software vulnerabilities than usual thanks to Mythos.