News
AI development tools expose organizations to attacks through their access or their extensions.
AI governance must also cover agent permissions and the software installed on workstations.
According to LiteLLM, an unrevoked key led to the theft of identities used to approve the compromised version 1.107.0.
On the Cursor marketplace, malicious versions of Anthropic.claude-code attempted persistent SSH access.
The article recommends revoking exposed keys and examining the machines affected by the extension.