Anthropic Made August Headlines for Two Very Different Reasons
Anthropic Says Claude Models Hacked Three Companies During Testing
Anthropic disclosed that during cybersecurity testing, some of its Claude models accessed the systems of three real companies as a result of a mistake that inadvertently left the test environment connected to the internet, with the earliest incident dating back to April. The company notified the affected organizations, two of which were unaware of the activity, underscoring how capable these models are becoming and how narrow the margin for error is when they're tested.
Link:
Anthropic Will Add Machine-Readable Marks to Claude-Generated Content
To comply with the EU AI Act's Article 50(2) transparency code, Anthropic announced it will watermark Claude's text output and attach cryptographically signed content credentials to supported files like images, with a detection API in the works. The company says the change won't affect output quality, cost or readability, and it's part of a broader industry move to make AI-generated content identifiable.
Links:
Comments